JPEG Column: 105th JPEG Meeting in Berlin, Germany

JPEG Trust becomes an International Standard

The 105th JPEG meeting was held in Berlin, Germany, from October 6 to 11, 2024. During this JPEG meeting, JPEG Trust was sent for publication as an International Standard. This is a major achievement in providing standardized tools to effectively fight against the proliferation of fake media and disinformation while restoring confidence in multimedia information.

In addition, the JPEG Committee also sent for publication the JPEG Pleno Holography standard, which is the first standardized solution for holographic content coding. This type of content might be represented by huge amounts of information, and efficient compression is needed to enable reliable and effective applications.

The following sections summarize the main highlights of the 105th JPEG meeting:

105th JPEG Meeting, held in Berlin, Germany.
  • JPEG Trust
  • JPEG Pleno
  • JPEG AI
  • JPEG XE
  • JPEG AIC
  • JPEG DNA
  • JPEG XS
  • JPEG XL


JPEG Trust

In an important milestone, the first part of JPEG Trust, the “Core Foundation” (ISO/IEC IS 21617-1) International Standard, has now been approved by the international ISO committee and is being published. This standard addresses the problem of dis- and misinformation and provides leadership in global interoperable media asset authenticity. JPEG Trust defines a framework for establishing trust in digital media.

Users of social media are challenged to assess the trustworthiness of the media they encounter, and agencies that depend on the authenticity of media assets must be concerned with mistaking fake media for real, with risks of real-world consequences. JPEG Trust provides a proactive approach to trust management. It is built upon and extends the Coalition for Content Provenance and Authenticity (C2PA) engine. The first part defines the JPEG Trust framework and provides building blocks for more elaborate use cases via its three main pillars:

  • Annotating provenance – linking media assets together with their associated provenance annotations in a tamper-evident manner
  • Extracting and evaluating Trust Indicators – specifying how to extract an extensive array of Trust Indicators from any given media asset for evaluation
  • Handling privacy and security concerns – providing protection for sensitive information based on the provision of JPEG Privacy and Security (ISO/IEC 19566-4)

Trust in digital media is context-dependent. JPEG Trust does NOT explicitly define trustworthiness but rather provides a framework and tools for proactively establishing trust in accordance with the trust conditions needed. The JPEG Trust framework outlined in the core foundation enables individuals, organizations, and governing institutions to identify specific conditions for trustworthiness, expressed in Trust Profiles, to evaluate relevant Trust Indicators according to the requirements for their specific usage scenarios. The resulting evaluation can be expressed in a Trust Report to make the information easily accessed and understood by end users.

JPEG Trust has an ambitious schedule of future work, including evolving and extending the core foundation into related topics of media tokenization and media asset watermarking, and assembling a library of common Trust Profile requirements.

JPEG Pleno

The JPEG Pleno Holography activity reached a major milestone with the FDIS of ISO/IEC 21794-5 being accepted and the International Standard being under preparation by ISO. This is a major achievement for this activity and is the result of the dedicated work of the JPEG Committee over a number of years. The JPEG Pleno Holography activity continues with the development of a White Paper on JPEG Pleno Holography to be released at the 106th JPEG meeting and planning for a workshop for future standardization on holography intended to be conducted in November or December 2024.

The JPEG Pleno Light Field activity focused on the 2nd edition of ISO/IEC 21794-2 (“Plenoptic image coding system (JPEG Pleno) Part 2: Light field coding”) which will integrate AMD1 of ISO/IEC 21794-2 (“Profiles and levels for JPEG Pleno Light Field Coding”) and include the specification of the third coding mode entitled Slanted 4D Transform Mode and the associated profile.

Following the Call for Contributions on Subjective Light Field Quality Assessment and as a result of the collaborative process, the JPEG Pleno Light Field is also preparing standardization activities for subjective and objective quality assessment of light fields. At the 105th JPEG meeting, collaborative subjective results on light field quality assessments were presented and discussed. The results will guide the subjective quality assessment standardization process, which has issued its fourth Working Draft.

The JPEG Pleno Point Cloud activity released a White Paper on JPEG Pleno Learning-based Point Cloud Coding. This document outlines the context, motivation, and scope of the upcoming Part 6 of ISO/IEC 21794 scheduled for publication in early 2025, as well as giving the basis of the new technology, use cases, performance, and future activities. This activity focuses on a new exploration study into the latent space optimization for the current Verification Model.

JPEG AI

At the 105th meeting JPEG AI activity primarily concentrated on advancing Part 2 (Profiling), Part 3 (Reference Software), and Part 4 (Conformance). Part 4 moved forward to the Committee Draft (CD) stage, while Parts 2 and 3 are anticipated to reach DIS at the next meeting. The conformance CD outlines different types of conformances: 1) strict conformance for decoded residuals; 2) soft conformance for decoded feature tensors, allowing minor deviations; and 3) soft conformance for decoded images, ensuring that image quality remains comparable to or better than the quality offered by the reference model. For decoded images, two types of soft conformance were introduced based on device capabilities. Discussions on Part 2 examined memory requirements for various JPEG AI VM codec configurations. Additionally, three core experiments were established during this meeting, focusing on JPEG AI subjective assessment, integerization, and the study of profiles and levels.

JPEG XE

The JPEG XE activity is currently focused on preparing for handling the open Final Call for Proposals on lossless coding of events. This activity revolves around a new and emerging image modality created by event-based visual sensors. JPEG XE is about the creation and development of a standard to represent events in an efficient way allowing interoperability between sensing, storage, and processing, targeting machine vision and other relevant applications. The Final Call for Proposals ends in March of 2025 and aims to receive relevant coding tools that will serve as a basis for a JPEG XE standard. The JPEG Committee is also preparing discussions on lossy coding of events and how to evaluate such lossy coding technologies in the future. The JPEG Committee invites those interested in JPEG XE activity to consider the public documents, available on jpeg.org. The Ad-hoc Group on event-based vision was re-established to continue work towards the 106th JPEG meeting. To stay informed about this activity, please join the event-based vision Ad-hoc Group mailing list.

JPEG AIC

Part 3 of JPEG AIC (AIC-3) advanced to the Committee Draft (CD) stage during the 105th JPEG meeting. AIC-3 defines a methodology for subjective assessment of the visual quality of high-fidelity images. Based on two test protocols—Boosted Triplet Comparisons and Plain Triplet Comparisons—it reconstructs a fine-grained quality scale in JND (Just Noticeable Difference) units. According to the defined work plan, JPEG AIC-3 is expected to advance to the Draft International Standard (DIS) stage by April 2025 and become an International Standard (IS) by October 2026. During this meeting, the JPEG Committee also focused on the upcoming Part 4 of JPEG AIC, which refers to the objective quality assessment of high-fidelity images.

JPEG DNA

JPEG DNA is an initiative aimed at developing a standard capable of representing bi-level, continuous-tone grey-scale, continuous-tone colour, or multichannel digital samples in a format using nucleotide sequences to support DNA storage. The JPEG DNA Verification Model was created during the 102nd JPEG meeting based on the performance assessments and descriptive analyses of the submitted solutions to the Call for Proposals, published at the 99th JPEG meeting. Several core experiments are continuously conducted to validate and improve this Verification Model (VM), leading to the creation of the first Working Draft of JPEG DNA during the 103rd JPEG meeting. At the 105th JPEG meeting, the committee created a New Work Item Proposal for JPEG DNA to make it an official ISO work item. The proposal stated that JPEG DNA would be a multi-part standard: Part 1—Core Coding System, Part 2—Profiles and Levels, Part 3—Reference Software, and Part 4—Conformance. The committee aims to reach the IS stage for Part 1 by April 2026.

JPEG XS

The third editions of JPEG XS, Part 1 – Core coding tools, Part 2 – Profiles and buffer models, and Part 3 – Transport and container formats, have now been published and made available on ISO. The JPEG Committee is finalizing the third edition of the remaining two parts of the JPEG XS standards suite, Part 4 – Conformance testing and Part 5 – Reference software. The FDIS of Party 4 was issued for the ballot at this meeting. Part 5 is still at the Committee Draft stage, and the DIS is planned for the next JPEG meeting. The reference software has a feature-complete decoder fully compliant with the 3rd edition. Work on the TDC profile encoder is ongoing.

JPEG XL

A third edition of JPEG XL Part 2 (File Format) will be initiated to add an embedding syntax for ISO 21496 gain maps, which can be used to represent a custom local tone mapping and have artistic control over the SDR rendition of an HDR image coded with JPEG XL. Work on hardware and software implementations continues, including a new Rust implementation.

Final Quote

“In its commitment to tackle dis/misinformation and to manage provenance, authorship, and ownership of multimedia information, the JPEG Committee has reached a major milestone by publishing the first ever ISO/IEC endorsed specifications for bringing back trust into multimedia. The committee will continue developing additional enhancements to JPEG Trust. New parts of the standard are under development to define a set of additional tools to further enhance interoperable trust mechanisms in multimedia.” said Prof. Touradj Ebrahimi, the Convenor of the JPEG Committee.

SIGMM Workshop on Multimodal AI Agents

The SIGMM Workshop on Multimodal AI Agents was held on October 28th, 2024, at ACMMM24 in Melbourne as an invitation-only event. The initiative was launched by Alberto Del Bimbo, Ramesh Jain, and Alan Smeaton following a vision of the future where multimedia expertise converges with the power of large language models and the belief that there is a great opportunity to position the Multimedia research community at the center of this transformation. The event was structured as three roundtables, inviting some of the most influential figures in the multimedia field to brainstorm on key issues. The goal was to design the future, identifying the multimodal opportunity in the days of powerful large-model systems and preparing an agenda for the coming years for the SIGMM community. We did not want to overlap with the current thinking of how multimodality will be included in the emerging large-models.  Instead, the goal was on how deep multimodality is essential in building next stages of AI agents for real world applications and how fundamental it is in understanding real-time contexts and for actions by agents. The event received a great response, with over 30 attendees from both Academia and Industry, representing 13 different countries.

Three roundtables focused on Tech ChallengesApplications, and Industry-University collaboration. The participants were divided into three groups and assigned to the three roundtables according to their profiles and preferences. For the roundtables, we did not prepare specific questions but rather outlined key areas of focus for discussion. A brief document that provided a short introduction for each roundtable, summarizing the topic of the debate and highlighting three major subjects to guide the discussion was prepared and given to the discussant a few days before the meeting. 

In the following we report a brief synthesis of the discussions at the roundtables, highlighting the principal arguments of discussion and proposals. 

Tech challenges Roundtable

Motivations for the discussion: As large pre-trained models become more prevalent and move towards multimodality, looking at the future, a key issue for their usage arises around the impact of their updating and fine-tuning, understanding how to ensure that improvements in one area don’t come at the cost of degradation in others. It is also fundamentally important to understand how deep multimodality is essential for building next stages of AI agents for real world applications, as well as for comprehending real-time contexts and guiding actions by agents towards Artificial General Intelligence. 

Some salient sentences, open questions, proposals from the discussion:

  • The interplay between human intelligence and machine intelligence is a fundamental aspect of what should be multi-modal. There are not yet deep enough multimodal models…. models for information that truly span all, or even a subset of modalities. We need metrics for this human-machine, human-intelligence machine-intelligence, action. We should come up with and define a task around how people collaborate productively. We should look at something like dynamic difficulty adjustment, that requires continuous, real-time development or training. 
  • Benchmarks are of crucial importance, not just to evaluate one thing against another thing, but to stretch the capabilities. It is not just about passing the benchmark; it is about setting the targets. We should envision a SIGMM-endorsed or sponsored multimodal benchmark by approaching some big tech companies to benchmark some multimodal activity within and across companies.

Applications Roundtable

Motivations for the discussion:   Multimodality is a cornerstone of emerging real-world applications, providing context and situational awareness to systems. Large Multimodal Models are credited for transforming various industries and enabling new applications. Key challenges lie in developing computational approaches for media fusion to construct context and situational understanding, addressing real-time computing costs, and refining model building. It is therefore essential for the SIGMM community to reason on how to build a vibrant community around one or a few key applications.

Some salient sentences, open questions, proposals from the discussion:

There are many areas for application where the SIGMM community can provide vital and innovative contributions and should concentrate its applicative research. Example application areas  and examples of research are: 

  • Health: there is an absence of open-ended sensory data representing of long-term complex information in the health area. We can think of integrated, federated machine learning, i.e. an integrated, federated data space for data control. 
  • Education: we can think of some futuristic learning approach, like completely autonomous learning.  Namely, AI agents that will be supportive through observation models, able to adjust the learning level so that some can finish faster than the others and learn depending on the modalities they like to receive. It is also of key importance to consider what the role of teacher and the role of AI is. 
  • Productivitywe can think of tools for immersive multi-modal experiences, to generate cross-modal content including 3D and podcasting in immersive environments.
  • Entertainment: we should think of how we can improve entertainment through immersive story driven experiences. 

Industry and University Roundtable

Motivations for the discussion:   Research on large AI models is by far dominated by private companies, thanks in part to their access to the data and the cost for building and training such models. As a result, academic institutions are being left behind in the AI race. It is therefore urgent to reason about which research directions are viable for universities and think of new Industry-University collaboration models for multimodal AI research.  It is also important to capitalize on the unique advantage of Academy, concerning their neutrality and ability to address long-term social and ethical issues related to technology.

Some salient sentences, open questions, proposals from the discussion:

  • Small and medium enterprises feel that they are left out. These are the ones who came to talk to universities. This is an opportunity for the SIGMM community to see how we can help.  SIGMM could sponsor joint PhD programs for example addressing small size, multi-model, foundation models, or intelligent agents, where a company sponsors part of the grant project. 
  • SIGMM should promote large visibility events at ACM Multimedia like Grand Challenges and Hackathons. As a community we could sponsor a company-wise Grand Challenge on multimodal AI and intelligent agents, leveraging industry to contribute more data sets. We could promote a regional-global Hackathon where Hackathons are held and overseen in different regions in the world, and the top teams then invited to come to ACM Multimedia and compete for it. 

Based on the discussions at the roundtables, we have identified several concrete actions that could help position the SIGMM research community at the forefront of the multimodal AI transformation:

At the next ACM Multimedia Conference

  • Explicit inclusion of multimodality as a key topic in the next ACM Multimedia call.
  • Multimodal Hackathon on Intelligent Agents (regional-global hackathon).
  • Multimodal Benchmarks (collaborations within and across major tech companies).
  • Multimodal Grand Challenges (in partnership with industry leaders).

At the next ACM SIGMM call for Special projects

  • Special Projects focused on Multimodal AI.

SIGMM is committed to pursuing these initiatives.

Diversity and Inclusion in focus at ACM IMX 2024

Summary: ACM IMX 2024 took place in Stockholm, Sweden, from June 12 to 14, continuing its dedication to promoting diversity within the community. Recognising the importance of amplifying varied voices and experiences to advance the field, the conference built on prior achievements in diversity and inclusion of IMX through a series of initiatives to promote diversity and inclusion (D&I).  This column provides a concise overview of the main D&I initiatives, including childcare support, early-career researcher grants, and manuscript accessibility support.  It includes participant feedback and short testimonials shared during and after the conference to highlight the value of these initiatives. 

To encourage a broad and inclusive pool of organisers, one method employed by the general chairs of ACM IMX’24 to prioritise diversity and inclusion was to team seasoned committee members with new members within the organising committee, this was done as a method to actively foster mentoring opportunities that support continuity and the development of future conference leadership. In addition to this, IMX’24 invited community members to self-nominate for various chair and organisational roles to make it clear that chair roles were open and available to all who were interested in being part of organising the conference. This call for applications was announced during the closing session of ACM IMX’23 in Nantes, France and, over a two-month period, the committee received 12 applications from which 5 candidates were selected to serve as chairs in various capacities. This inclusive approach allowed ACM IMX to engage with junior members and volunteers who might not have been reached through traditional recruitment methods, pairing them with experienced team members to ensure that they were able to build their network within the community and their skills in conference organisation and management. 

SIGMM support was used to enable the chairs of IMX’24 to introduce several initiatives to ensure that all individuals, regardless of personal circumstances, could participate fully in the conference. These initiatives had openly announced calls to all eligible community members who wished to attend the conference in person in Stockholm but required financial assistance. To ensure a fair and thorough selection, the IMX’24 Diversity and Inclusion Chairs, in collaboration with the General Chairs, reviewed each of the applications to ensure that the widest range of support could be offered with the available funds. Applications were evaluated on a rolling basis to ensure that participants were able to organise their travel and visa arrangements without the added challenges of time pressure.

With this support from SIGMM, Diversity and Inclusion grants for IMX were made available for participants, covering:

  • Travel Support for Non-Students from Marginalised and Underrepresented Groups: This grant provided travel support for researchers who self-identified as marginalised or underrepresented within the ACM IMX community, particularly those from non-WEIRD (Western, Educated, Industrialised, Rich, Developed) countries who lacked other funding opportunities. Priority was given to early-career researchers (such as post-docs), and those needing financial assistance, to compliment existing SIGCHI and SIGMM student targeted travel grants. 
  • Childcare and Parental Support: This grant offered financial assistance to parents attending ACM IMX’24, subsidising childcare costs to enable broader participation and to cover expenses related to children’s travel, travel for a childcare companion, and on-site or arranged babysitting during the conference.
  • Disability and Carer Support: This grant aimed to support attendees on extended leave from work due to disability, parental responsibilities, or other personal circumstances. Recipients of this award also received a complementary free conference registration. 
  • Student Travel Awards: SIGMM also provided awards directly to students to support travel expences, enabling a broader range of participation and complimenting free registration offered for those students volunteering at the conference. 

The SIGMM’s special initiatives for diversity and inclusion enable IMX’24 to secure a keynote designed to foster a more inclusive dialogue. Delivered by artist Jake Elwes—a self described hacker, radical faerie, and researcher—the keynote focused on “queer artificial intelligence” and featured deepfake drag performers. Elwes’ work invited the attendees to reflect on who builds these systems, the intentions behind them, and how they can be reclaimed to envision and create different visions of a technology enhanced future.

In combination with support from SIGMM, a special workshop focused on engaging with research and researchers from Latin America as a region of interest was made possible through the generous backing of the SIGCHI Development Fund (SDF). This enabled researchers and workshop keynote speakers to participate in both the “IMX in Latin America – 2nd International Workshop” and attend the conference. A core objective was to increase diversity by broadening the IMX community through actively encouraging colleagues from Latin America to attend and contribute. This workshop also published it’s submissions as part of the ACM IMX’24 workshop proceedings in ICPS.

For the first time at ACM IMX, an external provider (TAPS) was hired to ensure accessibility of papers prior to publication. Finally, the conference offered a range of venue-focused diversity and inclusion initiatives, including the provision of all-gender bathrooms, pronoun badges, and approachable senior community members to support engagement. Care corner and tables were thoughtfully set up throughout the conference to provide attendees with free hygiene essentials such as masks, refreshers, hand sanitisers, sanitary pads and tampons. These measures highlighted ACM IMX’24 commitment to fostering a welcoming and accessible environment for all participants.

Figure 1: Participants’ responses on their perception of diversity and inclusion at IMX, highlighting that it encompasses representation, welcoming environments, active engagement, research focus, and shaping future media experiences.

“During the closing event of IMX2024, we asked our attendees to answer a few questions that could help plan future IMX conferences. We asked everyone to share what future research directions could be included to address D&I at IMX. Some of the suggestions were to include the field of Humanities, to study usability among different demographics, and to understand how people who might not have economic access to technology could benefit from such technology. We also asked everyone to select what, according to them, is D&I at IMX. The options Everyone feels welcomed, Diverse individuals are able to engage and contribute and People from diverse backgrounds get represented and have a voice received a majority of the votes when compared to “Shape the future of interactive media experiences and “Research that focuses on diversity and inclusion in media experiences”. When asked to share how included they felt at IMX2024, 92% of the participants shared that they either felt included or very much [with some leaving the question unanswered]. They also shared how different aspects made them feel included. Some of the highlights were the care corner that was arranged to support the basic needs of the attendees, the social events, interactions at the conference, and the community. ” – Sujithra Raviselvam, IMX’24 Diversity and Inclusion Co-Chair.

Figure 2: Participants’ feedback on factors contributing to feelings of inclusion and exclusion at IMX, along with suggestions for future research directions aimed at improving diversity and inclusion. The feedback highlights personal interactions, event organization, and amenities as key to feeling included, while future research suggestions focus on enhancing accessibility, providing economic support, and integrating more diverse perspectives in HCI research.

The best way to understand the impacts of these supports is through the words of those who were enabled to join the conference by receiving it. 

The grant received for IMX2024 allowed me to attend the conference. Having a young child is challenging as an early researcher, as you must, sometimes, sacrifice your career or family. This grant allowed me to travel without any of these. I could attend the conference without stress or second thoughts, and support my family during the few days of the conference. Thanks to this, I received valuable feedback on my work, followed interesting presentations, and did not miss my family.” – Romain Herault, childcare award recipient. 

“I had the opportunity to present our qualitative study focused on understanding the sensitive values of women entrepreneurs in Brazil to support designing multi-model conversational AI financial systems at IMX, followed by interesting discussions about it in the workshop organized by Debora Christina Muchaluat Saade, Mylene Farias and Jesus Favela. The conference was focused on the future of multimodal technologies, with many exciting demos to investigate, to make more accessible, and to challenge assumptions of real life through a multimedia lens. We also had a conference dinner with the theme of the midsummer celebration. I was amazed by its meaning; as far as I understood, the purpose is to celebrate the light, sun, and summer season with family and friends! I loved it! It was also an opportunity to explore the beautiful Stockholm city with new colleagues and meet current collaborators in research.”– Heloisa Caroline de Souza Pereira Candello.  

A total of 21 applicants received support through diversity and inclusion grants provided by both SIGMM and the SIGCHI Development Fund (SDF). This assistance enabled full participation in ACM IMX’24 and supported a diverse group, including students, non-students from marginalised backgrounds, early-career researchers, and Latin American researchers, all of whom benefitted from these grants and made up more than 10% of the total conference attendees – truly changing and undoubtedly enhancing the experience of all attendees at the conference. 

Figure 3: The word clouds present two data sets from an IMX survey: the countries respondents identify as home, and the locations they would like IMX to feature in the future. It highlights a diverse range of home countries, including Brazil, Germany, and India, and suggest future IMX locations such as Japan, Brazil, and various cities in the USA, indicating a global interest and the geographical diversity of the IMX community.

Reports from ACM Multimedia 2024

Introduction

The ACM Multimedia Conference 2024, held in Melbourne, Australia from October 28 to November 1, 2024, was a major event that brought together leading researchers, practitioners, and industry professionals in the field of multimedia. This year’s conference marked a significant milestone as it was the first time since the end of the COVID-19 pandemic that the event returned to the Asia-Pacific region and resumed as a fully in-person gathering. The event offered a dynamic platform for presenting cutting-edge research, exploring new trends, and fostering collaborations across academia and industry.

Held in Melbourne, a city known for its vibrant culture and technological advancements, the conference was well-organized, ensuring a seamless experience for all participants. As part of its ongoing commitment to supporting the next generation of multimedia researchers, SIGMM awarded Student Travel Grants to 24 students. Each recipient received up to 1,000 USD to cover their travel and accommodation expenses. These grants were intended to help students who showed academic promise but faced financial barriers, allowing them to fully engage with the conference and its events. To apply, students were required to submit an online form, and the selection committee chose the recipients based on academic excellence and demonstrated financial need.

To give a voice to the travel grant recipients, we interviewed several of them to hear about their experiences and the impact the conference had on their academic and professional development. Below are some of their reflections.

Zhedong Zhang – Hangzhou Dianzi University

ACM Multimedia 2024 in Melbourne was my first international academic conference, and I am incredibly grateful to SIGMM for providing the travel grant. It was a great honor to present my paper, “From Speaker to Dubber: Movie Dubbing with Prosody and Duration Consistency Learning”, and to receive the Best Paper Award. As a PhD student, this recognition means a lot to me and encourages me to keep pushing forward with my research. 

Beyond the academic presentations, I had the chance to meet many brilliant researchers and fellow PhD students. I made connections with scholars working on similar topics and exchanged ideas that will help improve my work. The networking events and social gatherings were also highlights, as they allowed me to build friendships with colleagues from different parts of the world. I am truly grateful to SIGMM for making this experience possible and for the chance to be part of such a vibrant and inspiring academic community. I look forward to continuing my research and contributing to this exciting field.

Wu Tao – Zhejiang University

I’m incredibly grateful to the SIGMM team for awarding me this student travel grant – it really helped me a lot. I got to learn about so many fascinating papers at the conference and meet some brilliant professors and students. I even see some potential for future collaborations. I also had the chance to meet some big names in the field, like Tat-Seng Chua, who I’ve admired for a while. Meeting him, chatting, and even taking a photo with him felt like a once-in-a-lifetime opportunity, and I’m so thankful for it.

As for my own paper, I was both surprised and thrilled to see it actually got quite a bit of attention. At the welcome reception on the second day – before the poster session even began and before I’d even put up my poster – I noticed a few students already looking it up on their laptops. During the poster session, which was supposed to be two hours but probably stretched to three, I had a steady stream of people coming by to check out my work and ask questions. Some people even approached me earlier that morning. It was incredibly motivating to feel that kind of recognition and interest in what I’m working on. Thank you once again for this generous support! I look forward to attending the conference again.

Jianjun Qiao (Southwest Jiaotong University)

Attending ACM Multimedia 2024 in Melbourne was an incredible opportunity that greatly enriched my academic journey. This was my first time participating in an in-person conference, and I’m so grateful for the experience. The keynotes were fascinating, especially the talk on the Multimodal LLMs, which has significantly influenced my current research. I also enjoyed the poster sessions, where I could present my own work and engage in meaningful discussions with researchers from diverse backgrounds. The networking opportunities were invaluable, and I made several connections that I believe will lead to fruitful collaborations. I would like to extend my sincere thanks to SIGMM for the travel grant, which made my attendance possible. It was truly an unforgettable experience.

Changli Wu (Xiamen University)

ACM Multimedia 2024 was an unforgettable experience that exceeded all my expectations. As a PhD student, this was my first time presenting my research on 3D Referring Expression Segmentation at such a prestigious conference. The discussions I had with other attendees were invaluable, and I received constructive feedback that will undoubtedly improve my work. The diversity of the sessions was a highlight for me, as I was exposed to a variety of multimedia topics that I hadn’t considered before. The conference also provided a unique opportunity to interact with industry leaders, and I am now considering how to apply my research in real-world settings. 

VQEG Column: VQEG Meeting July 2024

Introduction

The University of Klagenfurt (Austria) hosted from July 01-05, 2024 a plenary meeting of the Video Quality Experts Group (VQEG). More than 110 participants from 20 different countries could attend this meeting in person and remotely.

The first three days of the meeting were dedicated to presentations and discussions about topics related to the ongoing projects within VQEG, while during the last two days an IUT-T Study Group G12 Question 19 (SG12/Q9) interim meeting took place. All the related information, minutes, and files from the meeting are available online in the VQEG meeting website, and video recordings of the meeting are available in Youtube.

All the topics mentioned bellow can be of interest for the SIGMM community working on quality assessment, but special attention can be devoted to the workshop on quality assessment towards 6G held within the 5GKPI group, and to the dedicated meeting of the IMG group hosted by the Distributed and Interactive Systems Group (DIS) of the CWI in September 2024 to work on ITU-T P.IXC recommendation. In addition, during those days there was a co-located ITU-T SG12 Q19 interim meeting.

Readers of these columns interested in the ongoing projects of VQEG are encouraged to subscribe to their corresponding reflectors to follow the activities going on and to get involved in them.

Another plenary meeting of VQEG has taken place from 18th 22nd of November 2024 and will be reported in a following issue of the ACM SIGMM Records.

VQEG plenary meeting at University of Klagenfurt (Austria), from July 01-05, 2024

Overview of VQEG Projects

Audiovisual HD (AVHD)

The AVHD group works on developing and validating subjective and objective methods to analyze commonly available video systems. During the meeting, there were 8 presentations covering very diverse topics within this project, such as open-source efforts, quality models, and subjective assessment methodologies:

Quality Assessment for Health applications (QAH)

The QAH group is focused on the quality assessment of health applications. It addresses subjective evaluation, generation of datasets, development of objective metrics, and task-based approaches. Joshua Maraval and Meriem Outtas (INSA Rennes, France) a dual rig approach for capturing multi-view video and spatialized audio capture for medical training applications, including a dataset for quality assessment purposes.

Statistical Analysis Methods (SAM)

The group SAM investigates on analysis methods both for the results of subjective experiments and for objective quality models and metrics. The following presentations were delivered during the meeting:  

No Reference Metrics (NORM)

The group NORM addresses a collaborative effort to develop no-reference metrics for monitoring visual service quality. In this sense, the following topics were covered:

  • Yixu Chen (Amazon, US) presented their development of a metric tailored for video compression and scaling, which can extrapolate to different dynamic ranges, is suitable for real-time video quality metrics delivery in the bitstream, and can achieve better correlation than VMAF and P.1204.3.
  • Filip Korus (AGH University of Krakow, Poland) talked about the detection of hard-to-compress video sequences (e.g., video content generated during e-sports events) based on objective quality metrics, and proposed a machine-learning model to assess compression difficulty.
  • Hadi Amirpour (University of Klagenfurt, Austria) provided a summary of activities in video complexity analysis, covering from VCA to DeepVCA and describing a Grand Challenge on Video Complexity.
  • Pierre Lebreton (Capacités & Nantes Université, France) presented a new dataset that allows studying the differences among existing UGC video datasets, in terms of characteristics, covered range of quality, and the implication of these quality ranges on training and validation performance of quality prediction models.
  • Zhengzhong Tu (Texas A&M University, US) introduced a comprehensive video quality evaluator (COVER) designed to evaluate video quality holistically, from a technical, aesthetic, and semantic perspective. It is based on leveraging three parallel branches: a Swin Transformer backbone to predict technical quality, a ConvNet employed to derive aesthetic quality, and a CLIP image encoder to obtain semantic quality.

Emerging Technologies Group (ETG)

The ETG group focuses on various aspects of multimedia that, although they are not necessarily directly related to “video quality”, can indirectly impact the work carried out within VQEG and are not addressed by any of the existing VQEG groups. In particular, this group aims to provide a common platform for people to gather together and discuss new emerging topics, possible collaborations in the form of joint survey papers, funding proposals, etc. During this meeting, the following presentations were delivered:

Joint Effort Group (JEG) – Hybrid

The group JEG-Hybrid addresses several areas of Video Quality Assessment (VQA), such as the creation of a large dataset for training such models using full-reference metrics instead of subjective metrics. In addition, the group includes the VQEG project Implementer’s Guide for Video Quality Metrics (IGVQM). The chair of this group,  Enrico Masala (Politecnico di Torino, Italy) presented the updates on the latest activities going on, including the status of the IGVQM project and a new image dataset, which will be partially subjectively annotated, to train DNN models to predict single user’s subjective quality perception. In addition to this:

Immersive Media Group (IMG)

The IMG group researches on the quality assessment of immersive media technologies. Currently, the main joint activity of the group is the development of a test plan to evaluate the QoE of immersive interactive communication systems, which is carried out in collaboration with ITU-T through the work item P.IXC. In this meeting, Pablo Pérez (Nokia XR Lab, Spain) and Jesús Gutiérrez (Universidad Politécnica de Madrid, Spain) provided an update on the progress of the test plan, reviewing the status of the subjective tests that were being performed at the 13 involved labs. Also in relation with this test plan:

In relation with other topics addressed by IMG:

In addition, a specific meeting of the group was held at Distributed and Interactive Systems Group (DIS) of CWI in Amsterdam (Netherlands) from the 2nd to the 4th of September to progress on the joint test plan for evaluating immersive communication systems. A total of 26 international experts from seven countries (Netherlands, Spain, Italy, UK, Sweden, Germany, US, and Poland) participated, with 7 attending online. In particular, the meeting featured presentations on the status of tests run by 13 participating labs, leading to insightful discussions and progress towards the ITU-T P.IXC recommendation.

IMG meeting at CWI (2-4 September, 2024, Netherlands)

Quality Assessment for Computer Vision Applications (QACoViA)

The group QACoViA addresses the study the visual quality requirements for computer vision methods, where the final user is an algorithm. In this meeting, Mikołaj Leszczuk (AGH University of Krakow, Poland) presented a study introducing a novel evaluation framework designed to address accurately predicting the impact of different quality factors on recognition algorithm, by focusing on machine vision rather than human perceptual quality metrics.

5G Key Performance Indicators (5GKPI)

The 5GKPI group studies relationship between key performance indicators of new 5G networks and QoE of video services on top of them. In this meeting, a workshop was organized by Pablo Pérez (Nokia XR Lab, Spain) and Kjell Brunnström (RISE, Sweden) on “Future directions of 5GKPI: Towards 6G“.

The workshop consisted of a set of diverse topics such as: QoS and QoE management in 5G/6G networks by (Michelle Zorzi, University of Padova, Italy); parametric QoE models and QoE management by Tobias Hoßfeld (University of. Würzburb, Germany) and Pablo Pérez (Nokia XR Lab, Spain); current status of standardization and industry by Kjell Brunnström (RISE, Sweden) and Gunilla Berndtsson (Ericsson); content and applications provider perspectives on QoE management by François Blouin (Meta, US); and communications service provider perspectives by Theo Karagioules and Emir Halepovic (AT&T, US). In addition, a panel moderated by Narciso García (Universidad Politécnica de Madrid, Spain) with Christian Timmerer (University of Klagenfurt, Austria), Enrico Masala (Politecnico di Torino, Italy) and Francois Blouin (Meta, US) as speakers.

Human Factors for Visual Experiences (HFVE)

The HFVE group covers human factors related to audiovisual experiences and upholds the liaison relation between VQEG and the IEEE standardization group P3333.1. In this meeting, there were two presentations related to these topics:

  • Mikołaj Leszczuk and Kamil Koniuch (AGH University of Krakow, Poland) presented a two-part insight into the realm of image quality assessment: 1) it provided an overview of the TUFIQoE project (Towards Better Understanding of Factors Influencing the QoE by More Ecologically-Valid Evaluation Standards) with a focus on challenges related to ecological validity; and 2) it delved into the ‘Psychological Image Quality’ experiment, highlighting the influence of emotional content on multimedia quality perception.

MPEG Column: 148th MPEG Meeting in Kemer, Türkiye

The 148th MPEG meeting took place in Kemer, Türkiye, from November 4 to 8, 2024. The official press release can be found here and includes the following highlights:

  • Point Cloud Coding: AI-based point cloud coding & enhanced G-PCC
  • MPEG Systems: New Part of MPEG DASH for redundant encoding and packaging, reference software and conformance of ISOBMFF, and a new structural CMAF brand profile
  • Video Coding: New part of MPEG-AI and 2nd edition of conformance and reference software for MPEG Immersive Video (MIV)
  • MPEG completes subjective quality testing for film grain synthesis using the Film Grain Characteristics SEI message
148th MPEG Meeting, Kemer, Türkiye, November 4-8, 2024.

Point Cloud Coding

At the 148th MPEG meeting, MPEG Coding of 3D Graphics and Haptics (WG 7) launched a new AI-based Point Cloud Coding standardization project. MPEG WG 7 reviewed six responses to a Call for Proposals (CfP) issued in April 2024 targeting the full range of point cloud formats, from dense point clouds used in immersive applications to sparse point clouds generated by Light Detection and Ranging (LiDAR) sensors in autonomous driving. With bit depths ranging from 10 to 18 bits, the CfP called for solutions that could meet the precision requirements of these varied use cases.

Among the six reviewed proposals, the leading proposal distinguished itself with a hybrid coding strategy that integrates end-to-end learning-based geometry coding and traditional attribute coding. This proposal demonstrated exceptional adaptability, capable of efficiently encoding both dense point clouds for immersive experiences and sparse point clouds from LiDAR sensors. With its unified design, the system supports inter-prediction coding using a shared model with intra-coding, applicable across various bitrate requirements without retraining. Furthermore, the proposal offers flexible configurations for both lossy and lossless geometry coding.

Performance assessments highlighted the leading proposal’s effectiveness, with significant bitrate reductions compared to traditional codecs: a 47% reduction for dense, dynamic sequences in immersive applications and a 35% reduction for sparse dynamic sequences in LiDAR data. For combined geometry and attribute coding, it achieved a 40% bitrate reduction across both dense and sparse dynamic sequences, while subjective evaluations confirmed its superior visual quality over baseline codecs.

The leading proposal has been selected as the initial test model, which can be seen as a baseline implementation for future improvements and developments. Additionally, MPEG issued a working draft and common test conditions.

Research aspects: The initial test model, like those for other codec test models, is typically available as open source. This enables both academia and industry to contribute to refining various elements of the upcoming AI-based Point Cloud Coding standard. Of particular interest is how training data and processes are incorporated into the standardization project and their impact on the final standard.

Another point cloud-related project is called Enhanced G-PCC, which introduces several advanced features to improve the compression and transmission of 3D point clouds. Notable enhancements include inter-frame coding, refined octree coding techniques, Trisoup surface coding for smoother geometry representation, and dynamic Optimal Binarization with Update On-the-fly (OBUF) modules. These updates provide higher compression efficiency while managing computational complexity and memory usage, making them particularly advantageous for real-time processing and high visual fidelity applications, such as LiDAR data for autonomous driving and dense point clouds for immersive media.

By adding this new part to MPEG-I, MPEG addresses the industry’s growing demand for scalable, versatile 3D compression technology capable of handling both dense and sparse point clouds. Enhanced G-PCC provides a robust framework that meets the diverse needs of both current and emerging applications in 3D graphics and multimedia, solidifying its role as a vital component of modern multimedia systems.

MPEG Systems Updates

At its 148th meeting, MPEG Systems (WG 3) worked on the following aspects, among others:

  • New Part of MPEG DASH for redundant encoding and packaging
  • Reference software and conformance of ISOBMFF
  • A new structural CMAF brand profile

The second edition of ISO/IEC 14496-32 (ISOBMFF) introduces updated reference software and conformance guidelines, and the new CMAF brand profile supports Multi-View High Efficiency Video Coding (MV-HEVC), which is compatible with devices like Apple Vision Pro and Meta Quest 3.

The new part of MPEG DASH, ISO/IEC 23009-9, addresses redundant encoding and packaging for segmented live media (REAP). The standard is designed for scenarios where redundant encoding and packaging are essential, such as 24/7 live media production and distribution in cloud-based workflows. It specifies formats for interchangeable live media ingest and stream announcements, as well as formats for generating interchangeable media presentation descriptions. Additionally, it provides failover support and mechanisms for reintegrating distributed components in the workflow, whether they involve file-based content, live inputs, or a combination of both.

Research aspects: With the FDIS of MPEG DASH REAP available, the following topics offer potential for both academic and industry-driven research aligned with the standard’s objectives (in no particular order or priority):

  • Optimization of redundant encoding and packaging: Investigate methods to minimize resource usage (e.g., computational power, storage, and bandwidth) in redundant encoding and packaging workflows. Explore trade-offs between redundancy levels and quality of service (QoS) in segmented live media scenarios.
  • Interoperability of live media Ingest formats: Evaluate the interoperability of the standard’s formats with existing live media workflows and tools. Develop techniques for seamless integration with legacy systems and emerging cloud-based media workflows.
  • Failover mechanisms for cloud-based workflows: Study the reliability and latency of failover mechanisms in distributed live media workflows. Propose enhancements to the reintegration of failed components to maintain uninterrupted service.
  • Standardized stream announcements and descriptions: Analyze the efficiency and scalability of stream announcement formats in large-scale live streaming scenarios. Research methods for dynamically updating media presentation descriptions during live events.
  • Hybrid workflow support: Investigate the challenges and opportunities in combining file-based and live input workflows within the standard. Explore strategies for adaptive workflow transitions between live and on-demand content.
  • Cloud-based workflow scalability: Examine the scalability of the REAP standard in high-demand scenarios, such as global live event streaming. Study the impact of cloud-based distributed workflows on latency and synchronization.
  • Security and resilience: Research security challenges related to redundant encoding and packaging in cloud environments. Develop techniques to enhance the resilience of workflows against cyberattacks or system failures.
  • Performance metrics and quality assessment: Define performance metrics for evaluating the effectiveness of REAP in live media workflows. Explore objective and subjective quality assessment methods for media streams delivered using this standard.

The current/updated status of MPEG-DASH is shown in the figure below.

MPEG-DASH status, November 2024.

Video Coding Updates

In terms of video coding, two noteworthy updates are described here:

  • Part 3 of MPEG-AI, ISO/IEC 23888-3 – Optimization of encoders and receiving systems for machine analysis of coded video content, reached Committee Draft Technical Report (CDTR) status
  • Second edition of conformance and reference software for MPEG Immersive Video (MIV). This draft includes verified and validated conformance bitstreams and encoding and decoding reference software based on version 22 of the Test model for MPEG immersive video (TMIV). The test model, objective metrics, and some other tools are publicly available at https://gitlab.com/mpeg-i-visual.

Part 3 of MPEG-AI, ISO/IEC 23888-3: This new technical report on “optimization of encoders and receiving systems for machine analysis of coded video content” is based on software experiments conducted by JVET, focusing on optimizing non-normative elements such as preprocessing, encoder settings, and postprocessing. The research explored scenarios where video signals, decoded from bitstreams compliant with the latest video compression standard, ISO/IEC 23090-3 – Versatile Video Coding (VVC), are intended for input into machine vision systems rather than for human viewing. Compared to the JVET VVC reference software encoder, which was originally optimized for human consumption, significant bit rate reductions were achieved when machine vision task precision was used as the performance criterion.

The report will include an annex with example software implementations of these non-normative algorithmic elements, applicable to VVC or other video compression standards. Additionally, it will explore the potential use of existing supplemental enhancement information messages from ISO/IEC 23002-7 – Versatile supplemental enhancement information messages for coded video bitstreams – for embedding metadata useful in these contexts.

Research aspects: (1) Focus on optimizing video encoding for machine vision tasks by refining preprocessing, encoder settings, and postprocessing to improve bit rate efficiency and task precision, compared to traditional approaches for human viewing. (2) Examine the use of metadata, specifically SEI messages from ISO/IEC 23002-7, to enhance machine analysis of compressed video, improving adaptability, performance, and interoperability.

Subjective Quality Testing for Film Grain Synthesis

At the 148th MPEG meeting , the MPEG Joint Video Experts Team (JVET) with ITU-T SG 16 (WG 5 / JVET) and MPEG Visual Quality Assessment (AG 5) conducted a formal expert viewing experiment to assess the impact of film grain synthesis on the subjective quality of video content. This evaluation specifically focused on film grain synthesis controlled by the Film Grain Characteristics (FGC) supplemental enhancement information (SEI) message. The study aimed to demonstrate the capability of film grain synthesis to mask compression artifacts introduced by the underlying video coding schemes.

For the evaluation, FGC SEI messages were adapted to a diverse set of video sequences, including scans of original film material, digital camera noise, and synthetic film grain artificially applied to digitally captured video. The subjective performance of video reconstructed from VVC and HEVC bitstreams was compared with and without film grain synthesis. The results highlighted the effectiveness of film grain synthesis, showing a significant improvement in subjective quality and enabling bitrate savings of up to a factor of 10 for certain test points.

This study opens several avenues for further research:

  • Optimization of film grain synthesis techniques: Investigating how different grain synthesis methods affect the perceptual quality of video across a broader range of content and compression levels.
  • Compression artifact mitigation: Exploring the interaction between film grain synthesis and specific types of compression artifacts, with a focus on improving masking efficiency.
  • Adaptation of FGC SEI messages: Developing advanced algorithms for tailoring FGC SEI messages to dynamically adapt to diverse video characteristics, including real-time encoding scenarios.
  • Bitrate savings analysis: Examining the trade-offs between bitrate savings and subjective quality across various coding standards and network conditions.

The 149th MPEG meeting will be held in Geneva, Switzerland from January 20-24, 2025. Click here for more information about MPEG meetings and their developments.

JPEG Column: 104th JPEG Meeting in Sapporo, Japan

JPEG XE issues Call for Proposals on event-based vision representation

The 104th JPEG meeting was held in Sapporo, Japan from July 15 to 19, 2024. During this JPEG meeting, a Call for Proposals on event-based vision representation was launched for the creation of the first standardised representation of this type of data. This CfP addresses lossless coding, and aims to provide the first standard representation for event-based data that ensures interoperability between systems and devices.

Furthermore, the JPEG Committee pursued its work in various standardisation activities, particularly the development of new learning-based technology codecs and JPEG Trust.

The following summarises the main highlights of the 104th JPEG meeting.

Event based vision reconstruction (from IEEE Spectrum, Feb. 2020).
  • JPEG XE
  • JPEG Trust
  • JPEG AI
  • JPEG Pleno Learning-based Point Cloud coding
  • JPEG Pleno Light Field
  • JPEG AIC
  • JPEG Systems
  • JPEG DNA
  • JPEG XS
  • JPEG XL

JPEG XE

The JPEG Committee continued its activity on JPEG XE and event-based vision. This activity revolves around a new and emerging image modality created by event-based visual sensors. JPEG XE is about the creation and development of a standard to represent events in an efficient way allowing interoperability between sensing, storage, and processing, targeting machine vision and other relevant applications. The JPEG Committee completed the Common Test Conditions (CTC) v2.0 document that provides the means to perform an evaluation of candidate technologies for efficient coding of events. The Common Test Conditions document also defines a canonical raw event format, a reference dataset, a set of key performance metrics and an evaluation methodology.

The JPEG Committee furthermore issued a Final Call for Proposals (CfP) on lossless coding for event-based data. This call marks an important milestone in the standardization process and the JPEG Committee is eager to receive proposals. The deadline for submission of proposals is set to March 31st of 2025. Standardization will start with lossless coding of events as this has the most imminent application urgency in industry. However, the JPEG Committee acknowledges that lossy coding of events is also a valuable feature, which will be addressed at a later stage.

Accompanying these two new public documents, a revised Use Cases and Requirements v2.0 document was also released to provide a formal definition for lossless coding of events that is used in the CTC and the CfP.

All documents are publicly available on jpeg.org. The Ad-hoc Group on event-based vision was re-established to continue work towards the 105th JPEG meeting. To stay informed about this activity please join the event-based vision Ad-hoc Group mailing list.

JPEG Trust

JPEG Trust provides a comprehensive framework for individuals, organizations, and governing institutions interested in establishing an environment of trust for the media that they use, and supports trust in the media they share. At the 104th meeting, the JPEG Committee produced an updated version of the Use Cases and Requirements for JPEG Trust (v3.0). This document integrates additional use cases and requirements related to authorship, ownership, and rights declaration. The JPEG Committee also requested a new Part to JPEG Trust, entitled “Media asset watermarking”. This new Part will define the use of watermarking as one of the available components of the JPEG Trust framework to support usage scenarios for content authenticity, provenance, integrity, labeling, and binding between JPEG Trust metadata and corresponding media assets. This work will focus on various types of watermarking, including explicit or visible watermarking, invisible watermarking, and implicit watermarking of the media assets with relevant metadata.

JPEG AI

At the 104th meeting, the JPEG Committee reviewed recent integration efforts, following the adoption of the changes in the past meeting and the creation of a new version of the JPEG AI verification model. This version reflects the JPEG AI DIS text and was thoroughly evaluated for performance and functionalities, including bitrate matching, 4:2:0 coding, region adaptive quantization maps, and other key features. JPEG AI supports a multi-branch coding architecture with two encoders and three decoders, allowing for six compatible combinations that have been jointly trained. The compression efficiency improvements range from 12% to 27% over the VVC Intra coding anchor, with decoding complexities between 8 to 215 kMAC/px.

The meeting also focused on Part 2: Profiles and Levels, which is moving to Committee Draft consultation. Two main concepts have been established: 1) the stream profile, defining a specific subset of the code stream syntax along with permissible parameter values, and 2) the decoder profile, specifying a subset of the full JPEG AI decoder toolset required to obtain the decoded image. Additionally, Part 3: Reference Software and Part 5: File Format will also proceed to Committee Draft consultation. Part 4 is significant as it sets the conformance points for JPEG AI compliance, and some preliminary experiments have been conducted in this area.

JPEG Pleno Learning-based Point Cloud coding

Learning-based solutions are the state of the art for several computer vision tasks, such as those requiring high-level understanding of image semantics, e.g., image classification, face recognition and object segmentation, but also 3D processing tasks, e.g. visual enhancement and super-resolution. Learning-based point cloud coding solutions have demonstrated the ability to achieve competitive compression efficiency compared to available conventional point cloud coding solutions at equivalent subjective quality. At the 104th meeting, the JPEG Committee instigated balloting for the Draft International Standard (DIS) of ISO/IEC 21794 Information technology — Plenoptic image coding system (JPEG Pleno) — Part 6: Learning-based point cloud coding. This activity is on track for the publication of an International Standard in January 2025. The 104th meeting also began an exploration into advanced point cloud coding functionality, in particular the potential for progressive decoding of point clouds.

JPEG Pleno Light Field

The JPEG Pleno Light Field effort has an ongoing standardization activity concerning a novel light field coding architecture that delivers a single coding mode to efficiently code light fields spanning from narrow to wide baselines. This novel coding mode is depth information agnostic resulting in significant improvement in compression efficiency. The first version of the Working Draft of the JPEG Pleno Part 2: Light Field Coding second edition (ISO/IEC 21794-2 2ED), including this novel coding mode, was issued during the 104th JPEG meeting in Sapporo, Japan.

The JPEG PLeno Model (JPLM) provides reference implementations for the standardized technologies within the JPEG Pleno framework, including the JPEG Pleno Part 2 (ISO/IEC 21794-2). Improvements to the JPLM have been implemented and tested, including the design of a more user-friendly platform.

The JPEG Pleno Light Field effort is also preparing standardization activities in the domains of objective and subjective quality assessment for light fields, aiming to address other plenoptic modalities in the future. During the 104th JPEG meeting in Sapporo, Japan, the collaborative subjective experiments aiming at exploring various aspects of subjective light field quality assessments were presented and discussed. The outcomes of these experiments will guide the decisions during the subjective quality assessment standardization process, which has issued its third Working Draft. A new version of a specialized tool for subjective quality evaluation, that supports these experiments, has also been released.

JPEG AIC

At its 104th meeting, the JPEG Committee reviewed results from previous Core Experiments that collected subjective data for fine-grained quality assessments of compressed images ranging from high to near-lossless visual quality. These crowdsourcing experiments used triplet comparisons with and without boosted distortions, as well as double stimulus ratings on a visual analog scale. Analysis revealed that boosting increased the precision of reconstructed scale values by nearly a factor of two. Consequently, the JPEG Committee has decided to use triplet comparisons in the upcoming AIC-3.

The JPEG Committee also discussed JPEG AIC Part 4, which focuses on objective image quality assessments for compressed images in the high to near-lossless quality range. This includes developing methods to evaluate the performance of such objective image quality metrics. A draft call for contributions is planned for January 2025.

JPEG Systems

At the 104th meeting Part 10 of JPEG Systems (ISO/IEC 19566-10), the JPEG Systems Reference Software, reached the IS stage. This first version of the reference software provides a reference implementation and reference dataset for the JPEG Universal Metadata Box Format (JUMBF, ISO/IEC 19566-5). Meanwhile, work is in progress to extend the reference software implementations of additional Parts, including JPEG Privacy and Security and JPEG 360.

JPEG DNA

JPEG DNA is an initiative aimed at developing a standard capable of representing bi-level, continuous-tone grey-scale, continuous-tone colour, or multichannel digital samples in a format using nucleotide sequences to support DNA storage. A Call for Proposals was published at the 99th JPEG meeting. Based on the performance assessments and descriptive analyses of the submitted solutions, the JPEG DNA Verification Model was created during the 102nd JPEG meeting. Several core experiments were conducted to validate this Verification Model, leading to the creation of the first Working Draft of JPEG DNA during the 103rd JPEG meeting.

The next phase of this work involves newly defined core experiments to enhance the rate-distortion performance of the Verification Model and its robustness to insertion, deletion, and substitution errors. Additionally, core experiments to test robustness against substitution and indel noise are conducted. A core experiment was also performed to integrate JPEG AI into the JPEG DNA VM, and quality comparisons have been carried out. A study on visual quality assessment of JPEG AI as an alternative to JPEG XL in the VM will be carried out.

In parallel, efforts are underway to improve the noise simulator developed at the 102nd JPEG meeting, enabling a more realistic assessment of the Verification Model’s resilience to noise. There is also ongoing exploration of the performance of different clustering and consensus algorithms to further enhance the VM’s capabilities.

JPEG XS

The core parts of JPEG XS 3rd edition were prepared for immediate publication as International Standards. This means that Part 1 of the standard – Core coding tools, Part 2 – Profiles and buffer models, and Part 3 – Transport and container formats, will be available before the end of 2024. Part 4 – Conformance testing is currently still under DIS ballot and it will be finalized in October 2024. At the 104th meeting, the JPEG Committee continued the work on Part 5 – Reference software. This part is currently at Committee Draft stage and the DIS is planned for October 2024. The reference software has a feature-complete decoder that is fully compliant with the 3rd edition. Work on the encoder is ongoing.

Finally, additional experimental results were presented on how JPEG XS can be used over 5G mobile networks for wireless transmission of low-latency and high quality 6K/8K 360 degree views with mobile devices and VR headsets. This work will be continued.

JPEG XL

Objective metrics results for HDR images were investigated (using among others the ColorVideoVDP metric), indicating very promising compression performance of JPEG XL compared to other codecs like AVIF and JPEG 2000. Both the libjxl reference software encoder and a simulated candidate hardware encoder were tested. Subjective experiments for HDR images are planned.

The second editions of JPEG XL Part 1 (Core coding system) and Part 2 (File format) are now ready for publication. The second edition of JPEG XL Part 3 (Conformance testing) has moved to the FDIS stage.

Final Quote

“The JPEG Committee has reached a new milestone by releasing a new Call for Proposals to code events. This call is aimed at creating the first International Standard to efficiently represent events, enabling interoperability between devices and systems that rely on event sensing.” said Prof. Touradj Ebrahimi, the Convenor of the JPEG Committee.

One benchmarking cycle wraps up, and the next ramps up: News from the MediaEval Multimedia Benchmark

Introduction

MediaEval, the Multimedia Evaluation Benchmark, has offered a yearly set of multimedia challenges since 2010. MediaEval supports the development of algorithms and technologies for analyzing, exploring and accessing information in multimedia data. MediaEval aims to help make multimedia technology a force for good in society and for this reason focuses on tasks with a human or social aspect. Benchmarking contributes in two ways to advancing multimedia research. First, by offering standardized definitions of tasks and evaluation data sets, it makes it possible to fairly compare algorithms and, in this way, track progress. If we can understand which types of algorithms perform better, we can more easily find ways (and the motivation) to improve them. Second, benchmarking helps to direct the attention of the research community, for example, towards new tasks that are based on real-world needs, or towards known problems for which more research is necessary to have a solution that is good enough for a real world application scenario.

The 2023 MediaEval benchmarking season culminated with the yearly workshop, which was held in conjunction with MMM 2024 (https://www.mmm2024.org) in Amsterdam, Netherlands. It was a hybrid workshop, which also welcomed online participants. The workshop kicked off with a joint keynote with MMM 2024. Yiannis Kompatsiaris, Information Technologies Institute, CERTH, on Visual and Multimodal Disinformation Detection. The talk covered the implications of multimodal disinformation online and the challenges that must be faced in order to detect it. The workshop featured an invited speaker, Adriënne Mendrik, CEO & Co-founder of Eyra, supporting benchmarks with the online Next platform. She talked about benchmark challenge design for science and how the Next platform is currently being used in the Social Sciences.

More information about the workshop can be found at https://multimediaeval.github.io/editions/2023/ and the proceedings were published at  https://ceur-ws.org/Vol-3658/ In the rest of this article, we provide an overview of the highlights of the workshop as well as an outlook to the next edition of MediaEval in 2025.  

Tasks at MediaEval

The MultimediaEval Workshop 2023 featured five tasks that focused on human and social aspects of multimedia analysis.

Three of the tasks required participants to combine or cross modalities or even consider new modalities. The Musti: Multimodal Understanding of Smells in Texts and Images task challenged participants to detect and classify smell references in multilingual texts and images from the 17th to the 20th century. They needed to identify whether a text and image evoked the same smell source, detect specific smell sources, and apply zero-shot learning for untrained languages. The remaining two tasks emphasized the social aspects of multimedia. In the NewsImages: Connecting Text and Images task, participants worked with a dataset of news articles and images, predicting which image accompanied a news article. This task aimed to explore cases in which there is a link between a text and an image that goes beyond the text being a literal description of what was pictured in the image. The Predicting Video Memorability task required participants to predict how likely videos were to be remembered, both short- and long-term, and to use EEG data to predict whether specific individuals would remember a given video, combining visual features and neurological signals. 

Two of the tasks focused on pushing forward video analysis, to be useful to support experts in carrying out their jobs. The task SportsVideo: Fine-Grained Action Classification and Position Detection task strives to develop technology that will support coaches. To address this task, participants analyzed videos of table tennis and swimming competitions, detecting athlete positions, identifying strokes, classifying actions, and recognizing game events such as scores and sounds. The task Transparent Tracking of Spermatozoa strived to develop technology that will support medical professionals. Task participants were asked to track sperm cells in video recordings to evaluate male reproductive health. This involved localizing and tracking individual cells in real time, predicting their motility, and using bounding box data to assess sperm quality. The task emphasized both accuracy and processing efficiency, with subtasks involving graph data structures for motility prediction. 

Impressions of Student Participants

MediaEval is grateful to SIGMM for providing funding for three students who attended the MediaEval Workshop and greatly helped us with the organization of this edition: Iván Martín-Fernández and Sergio Esteban-Romero from Speech Technology and Machine Learning Group (GTHAU) – Universidad Politécnica de Madrid, and Xiaomeng Wang from Radboud University. Below the students provide their comments and impressions of the workshop.

“As a novel PhD student, I greatly valued my experience attending MediaEval 2023. I participated as the main author and presented work from our group, GTHAU – Universidad Politécnica de Madrid, on the Predicting Video Memorability Challenge. The opportunity to meet renowned colleagues and senior researchers, and learn from their experiences, provided valuable insight into what academia looks like from the inside. 

MediaEval offers a range of multimedia-related tasks, which may sometimes seem under the radar but are crucial in developing real-world applications. Moreover, the conference distinguishes itself by pushing the boundaries, going beyond just presenting results to foster a deeper understanding of the challenges being addressed. This makes it a truly enriching experience for both newcomers and seasoned professionals alike. 

Having volunteered and contributed to organizational tasks, I also gained first-hand insight into the inner workings of an academic conference, a facet I found particularly rewarding. Overall, MediaEval 2023 proved to be an exceptional blend of scientific rigor, collaborative spirit, and practical insights, making it an event I would highly recommend for anyone in the multimedia community.”

Iván Martín-Fernández, PhD Student, GTHAU – Universidad Politécnica de Madrid

“Attending MediaEval was an invaluable experience that allowed me to connect with a wide range of researchers and engage in discussions about the latest advancements in Artificial Intelligence. Presenting my work on the Multimedia Understanding of Smells in Text and Images (MUSTI) challenge was particularly insightful, as the feedback I received sparked ideas for future research. Additionally, volunteering and assisting with organizational tasks gave me a behind-the-scenes perspective on the significant effort required to coordinate an event like MediaEval. Overall, this experience was highly enriching, and I look forward to participating and collaborating in future editions of the workshop.”

Sergio Esteban-Romero, PhD Student, GTHAU – Universidad Politécnica de Madrid

“I was glad to be a student volunteer at MediaEval 2024. Collaborating with other volunteers, we organized submission files and prepared the facilities. Everyone was exceptionally kind and supportive.
In addition to volunteering, I also participated in the workshop as a paper author. I submitted a paper to the NewsImage task and delivered my first oral presentation. The atmosphere was highly academic, fostering insightful discussions. And I received valuable suggestions to improve my paper.  I truly appreciate this experience, both as a volunteer and as a participant.”

Xiaomeng Wang PhD Student, Data Science – Radboud University

Outlook to MediaEval 2025 

We are happy to announce that in 2025 MediaEval will be hosted in Dublin, Ireland, co-located with CBMI 2025. The Call for Task Proposals is now open, and details regarding submitting proposals can be found here: https://multimediaeval.github.io/2024/09/24/call.html. The final deadline for submitting your task proposals is Wed. 22nd January 2025. We will publish the list of tasks offered in March and registration for participation in MediaEval 2025 will open in April 2025.

For this edition of MediaEval we will again emphasize our “Quest for Insight”: we push beyond improving evaluation scores to achieving deeper understanding about the challenges, including data and the strengths and weaknesses of particular types of approaches, with the larger aim of understanding and explaining the concepts that the tasks revolve around, promoting reproducible research, and fostering a positive impact on society. We look forward to welcoming you to participate in the new benchmarking year.

Report from CBMI 2024

The 21st International Conference on Content-based Multimedia Indexing (CBMI) was hosted by Reykjavik University in cooperation with ACM, SIGMM, VTT and IEEE. The three-day event took place on September 20-22 in Reykjavik, Iceland. Like the year before, it was as an exclusively in-person event. Despite the remote location, an active volcano and in person attendance requirement, we are pleased to report that we had a perfect attendance of presenting authors. CBMI was started in France and still has strong European roots. Looking at the nationality of the submitting authors we can see 17 unique nationalities, 14 countries in Europe, 2 in Asia and 1 in North America.

Conference highlights

Figure 1: First keynote speaker being introduced.

Key elements of a successful conference are the keynote sessions. The first and opening keynote, titled “What does it mean to ‘work as intended’?” was presented by Dr. Cynthia C. S. Liem on day 1. In this talk Cynthia raised important questions on how complex it can be to define, measure and evaluate human-focused systems. Using real-world examples, she demonstrated how recently developed systems, that passed the traditional evaluating metrics, still failed when deployed in the real-world. Her talk was an important reminder that certain weaknesses in human-focused systems are only revealed when exposed to reality.

Figure 2: Keynote speaker Ingibjörg Jónsdóttir (left) and closing keynote speaker Hannes Högni Vilhjálmsson (right).

Traditionally there are only two keynotes at CBMI, first on day 1 and second on day 2. However, our planned second keynotes could not attend until the last day and thus a 3rd “surprise” keynote was organized on day 2 with the title “Remote Sensing of Natural Hazards”.  The speaker was Dr. Ingibjörg Jónsdóttir, an associate professor of geology at the University of Iceland. She gave a very interesting talk about the unique geology of Iceland, the threats posed by natural hazards and her work using remote sensing to monitor both sea ice and volcanoes. This talk was well received by attendees as it gave insight into the host country, the volcanic eruption that ended just a week before the start of the conference (7th in past 2 years on the Reykjanes Peninsula). This subject is highly relevant to community, as the analysis and prediction is based on multimodal data.

The planned second keynote took place in the last session on day 3 and was given by Dr. Hannes Högni Vilhjálmsson, professor at Reykjvik University. The talk, titled “Being Multimodal: What Building Virtual Humans has Taught us about Multimodality”, gave the audience a deep dive into lessons learnt from his 20+ years of experience of developing intelligent virtual agents with face-to-face communication skills. “I will review our attempts to capture, understand and analyze the multi-modal nature of human communication, and how we have built and evaluated systems that engage in and support such communication.” is a direct quote from his abstract of the talk. 

CBMI is a relatively small, but growing, conference that is built on a strong legacy and has a highly motivated community behind it. The special sessions have long played an important role at CBMI and this year there were 8 special sessions accepted.

  • AIMHDA: Advances in AI-Driven Medical and Health Data Analysis
  • CB4AMAS: Content-based Indexing for audio and music: from analysis to synthesis
  • ExMA: Explainability in Multimedia Analysis
  • IVR4B: Interactive Video Retrieval for Beginners
  • MAS4DT: Multimedia analysis and simulations for Digital Twins in the construction domain
  • MmIXR: Multimedia Indexing for XR
  • MIDRA: Multimodal Insights for Disaster Risk Management and Applications
  • UHBER: Multimodal Data Analysis for Understanding of Human Behaviour, Emotions and their Reasons
Figure 3: SS UHBER chair Dr.  E. Vildjunaite with a conference participant. 

The number of papers per session ranged from 2 to 8. The larger sessions (CB4AMAS, MmIXR and UHBER) used a discussion panel format that created a more inclusive atmosphere and, at times, sparked lively discussions. 

Figure 4: Images from the poster session and the IVR4B competition.

Especially popular with attendees was the competition that took place in the Interactive Video Retrieval for Beginners (IVR4B) session. This session was hosted right after the poster session in the wide open space of Reykjavik University’s foyer. 

Awards

The selection committee was unanimous in that the contribution of Lorenzo Bianchi, Fabio Carrara, Nicola Messina & Fabrizio Falchi, titled “Is CLIP the main roadblock for fine-grained open-world perception?”, was the best paper award winner. With the generous support of ACM SIGMM, they were awarded 500 Euros. As the best paper was indeed also a student paper, it was decided to also give the runner-up a 300 Euro award. The runner-up was the contribution of Recep Oguz Araz, Dmitry Bogdanov, Pablo Alonso-Jimenez and Frederic Font, titled “Evaluation of Deep Audio Representations for Semantic Sound Similarity”.

The best demonstration was awarded to Joshua David Springer, Gylfi Thor Gudmundsson and Marcel Kyas for “Lowering Barriers to Entry for Fully-Integrated Custom Payloads on a DJI Matrice”. 

The top two systems in the IVAR4B competition were also recognized: the first place was for Nick Pantelidis, Maria Pegia, Damianos Galanopoulos, et al. for “VERGE: Simplifying Video Search for Novice”; and the second place was for Giuseppe Amato, Paolo Bolettieri, Fabio Carrara, et al. for “VISIONE 5.0: toward evaluation with novice users”. 

Social events

The first day of the conference was quite eventful as before the poster and IVAR4B sessions Francois Pineau-Benois and Raphael Moraly of the Odyssée Quartet performed selected classical works in the “Music-meets-Science” cultural event. The goals of the latter are to bring live classical music content to the community of Multimedia Research. Musicians played a concert and then discussed with researchers, specifically involved into music analysis and retrieval. Such kind of exchanges between content creators and content analysis, indexing and retrieval researchers has been a distinctive feature of CBMI since 2018. 
This event would not have been possible without the generous support of ACM SIGMM.

The second day was no less entertaining as before the banquet attendees took a virtual flight over Iceland’s beautiful landscape via the services of FlyOver Iceland. 
The next CBMI’2025 will be hold in Dublin organized by DCU.

The 2nd Edition of International Summer School on Extended Reality Technology and Experience (XRTX)

The ACM Special Interest  Group on Multimedia (ACM SIGMM), co-sponsored the second edition of the International Summer School on Extended Reality Technology and Experience (XRTX), which was took place from July 8-11, 2024 in Madrid (Spain) hosted by Universidad Carlos III de Madrid. As in the first edition in 2023, in the organization also participated Universidad Politécnica de Madrid, Nokia, Universidad de Zaragoza and Universidad Rey Juan Carlos. It attracted 29 participants from different disciplines (e.g., engineering, computer science, psychology, etc.) and 7 different countries.

Students and organizers of the Summer School XRTX (July 8-11, 2024, Madrid)

The support from ACM SIGMM permitted to bring top researchers in the field of XR to deliver keynotes in different topics related to technological (e.g., volumetric video, XR for healthcare) and user-centric factors (e.g., user experience evaluation, multisensory experiences, etc.), such as Pablo César (CWI, Netherlands), Anthony Steed (UCL, UK), Manuela Chessa (University of Genoa, Italy), Qi Sun (New York University), Diego Gutiérrez (Universidad de Zaragoza, Spain), and Marianna Obrist (UCL, UK). Also, an industry session was also included, led by Gloria Touchard (Nokia, Spain).

Keynote sessions

In addition to these 7 keynotes, the program included a creative design session led by Elena Márquez-Segura (Universidad Carlos III, Spain), a tutorial on Ubiq given by Sebastian Friston (UCL, UK), and 2 practical sessions led by Telmo Zarraonandia (Universidad Carlos III, Spain) and Sandra Malpica (Universidad de Zaragoza, Spain) to get hands-on experience working with Unity for VR development .

Design and practical sessions

Moreover, ​there were poster and demo sessions distributed along the whole duration of the school, in which the participants could showcase their works.

Poster and demo sessions

The summer school was also sponsored by Nokia and the Computer Science and Engineering Department of Universidad Carlos III, which allowed to offer grants to support a number of students with the registration and traveling costs.

Finally, in addition to science, there was time for fun with social activities, like practicing real and VR archery and visiting and immersive exhibition about Pompeii.

Practicing real and VR archery
Immersive exhibition about Pompeii

The list of talks was:

  • “Towards volumetric video conferencing” by Pablo Cesar.
  • “Strategies for Designing and Evaluating eXtended Reality Experiences” by Anthony Steed.
  • “XR for healthcare: Immersive and interactive technologies for serious games and exergames”, by Manuela Chessa.
  • “Toward Human-Centered XR: Bridging Cognition and Computation”, by Qi Sun.
  • “Improving the user’s experience in VR” by Diego Gutiérrez.
  • “Why XR is important for Nokia”, Gloria Touchard.
  • “The Role of Multisensory Experiences in Extended Reality: Unlocking the Power of Smell” by Marianna Obrist.