4th Workshop on Image/Video/Audio Quality in Computer Vision and Generative AI

Home

Important Dates/Links:

Submission Deadline: 15 December, 2024 -> 21 December, 2024
Submission Link: https://cmt3.research.microsoft.com/ImageQuality2025
Acceptance Notification Deadline: 5 January, 2025
Camera Ready Papers Submission Deadline: 10 January, 2025, at 11:59 PM Pacific Daylight Time.
Workshop Day: 28 February, 2025
Workshop Website: https://wacv2025-image-quality-workshop2.github.io/index.html
Authors of accepted papers are required to present their work live, either in-person or remote. If pre-recorded videos are used, the authors are required to join the meeting online to do Q&A live.

Description:

Many machine learning tasks and computer vision algorithms are susceptible to image/video/audio quality artifacts. Nonetheless, most visual learning and vision systems assume high-quality image/video/audio as input. In reality, noises and distortions are common in image/video/audio capturing and acquisition process. Oftentimes, artifacts can be introduced in the video compression, transcoding, transmission, decoding, and/or rendering process. All of these quality issues play a critical role on the performance of learning algorithms, systems and applications, therefore could directly impact the customer experience.

Topics:

This workshop addresses topics related to image/video/audio quality in machine learning, computer vision, and generative AI. The topics include, but are not limited to:

Impact of image/video/audio quality in traditional machine learning and computer vision use cases such as object detection, segmentation, tracking, and recognition;
Analyze, model and learn the quality impact from image/video/audio acquisition, compression, transcoding, transmission, decoding, rendering, and/or display;
Techniques used to improve image/video/audio quality in terms of
- brightening, color adjustment, sharpening, inpainting, deblurring, denoising, de-hazing, de-raining, demosaicing,
- removing artifacts such as shadows, glare, and reflections, etc.,
- resolution, frame rate, color gamut, dynamic range (SDR vs. HDR), etc.,
- noise/echo cancellation, speech enhancement, etc.;
- Film grain preservation and synthesis.
Novel image/video/audio quality assessment methodologies: full reference, reduced-reference, and non-reference;
Impact of image/video/audio quality in multi-modal use cases;
Evaluate image/video/audio quality produced by generative AI;
Techniques to detect and mitigate audio/video synchronization issue;
Techniques to measure the quality consistency across different types of content in video (such as ads, movies, streamed content, etc.);
Subjective image/video/audio quality data creation, cleaning, and validation in both lab and crowd-sourcing setups;
Datasets, statistics, and theory of image/video/audio quality;
Research, applications and system development of the above.

Submission Guidelines and Review Process:

Authors are encouraged to submit high-quality, original (i.e., not been previously published or accepted for publication in substantially similar form in any peer-reviewed venue including journal, conference or workshop) research.
All submissions should follow the same template as for the main WACV2025 conference. Please refer to the WACV Author Kit (Overleaf template, ZIP Archieve).
The main paper has an 8-page limit, references do not count toward this. There is no limit on the number of pages in the supplementary material. Only pdf files are accepted.
Unlike the main conference(WACV2025), the review process for this workshop has only one round, and is single-blind. Authors do not have to be anonymized when submitting their work.
Please submit your paper via this link: https://cmt3.research.microsoft.com/ImageQuality2025
Authors of accepted papers will be notified via email by: 5 January, 2025

Competition

Overview

We are excited to announce the Cross Domain Logo Recognition Grand Challenge, to be hosted as part of the WACV 2025 workshop. This challenge aims to push the boundaries of computer vision and machine learning in tackling one of the most crucial yet challenging tasks in visual brand identity: logo recognition across diverse domains. Logo recognition plays a pivotal role in numerous industries, including brand management, marketing analytics, intellectual property protection, and e-commerce. The ability to accurately identify logos in various contexts is essential for monitoring brand presence, tracking marketing effectiveness, detecting counterfeit products, and enhancing user experiences in visual search applications. Despite its significance, logo recognition remains a formidable challenge in both academia and industry, particularly when dealing with large-scale, real-world scenarios. Logo detection is challenging due to the vast diversity in appearance, scale, perspective, and potential distortions or occlusions. These factors make it difficult for systems to consistently recognize logos across various products, environments, and viewing conditions. To further elevate the challenge, we are introducing a unique aspect that mirrors real-world scenarios: recognizing new logos based solely on their design templates . While humans can easily bridge this domain gap and identify logos on actual products after seeing only the original design, machine learning models struggle to generalize across such disparate domains. This aspect of the challenge simulates the practical need for systems that can quickly adapt to new brand identities without extensive real-world training data.

Challenge Introduction

To support this grand challenge, we will be releasing a comprehensive dataset that encompasses a wide range of logos across various product categories and real-world contexts. This dataset will serve as a benchmark for evaluating the performance of participating models in cross-domain logo recognition. Amazon (as one of the WACV sponsors) will provide cash prizes for the top 3 submissions. Cash prize:

First place: TBD
Second place: TBD
Third place: TBD

The prize awards will be granted based on performance over an unreleased test set. The selected winners will have an opportunity to present their work at the 4th Workshop on Image/Video/Audio Quality in Computer Vision and Generative AI

Dataset Details

The Cross-Domain one-shot Logo Recognition Dataset released for this challenge was created by Amazon and is released under the license CC BY SA 4.0 . For more information, please visit codabench.