ICCV 2023 Workshop on Resource Efficient Deep Learning for Computer Vision (RCV'23)

[Deepak Gupta]

ICCV 2023 Workshop on Resource-Efficient Deep Learning for Computer Vision (RCV'23)

Overview:

There exist several different directions towards making deep learning efficient for computer vision, thereby leading to a reduction in the required computational memory or the associated training and inference time. This workshop aims to bring together researchers and industry practitioners who work towards building efficient computer vision models with deep learning. It will also serve as a platform to discuss research efforts towards budget-aware model training and inference, as well as two challenges focusing on resource-efficient model training and inference. The topics that will be covered include budget-aware model training and inference, resource-efficient model training and inference, and more general topics from the domain of efficient deep learning. To push research efforts in this direction, we are also organizing two challenges focusing on resource-efficient model training and inference where participants would be required to optimize the model training process under computational memory constraint as well as the inference process under latency constraint.
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Important Dates
Workshop papers: Submission deadline: July 25, 2023 
⦁ Review starts: July 26, 2023
⦁ Review ends: Aug 03, 2023
⦁ Notification of accepted papers: Aug 05, 2023
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Workshop challenges:
⦁ Phase I starts: April 12, 2023
⦁ Phase II starts: May 10, 2023
⦁ Phase II ends: July 15, 2023
⦁ Final submission of code: July 17, 2023
⦁ Notification of results: July 25, 2023
⦁ Submission of report: Aug 05, 2023

This Call for Papers focuses on resource efficient deep learning for computer vision, and aims to bring together researchers and practitioners from academia and industry to share their expertise, ideas, and best practices. Topics of interest include designing efficient neural architectures, compression of existing models, efficient fine-tuning of large-scale vision and language models, deep learning on very large images, energy-efficient deep learning hardware and accelerators, benchmarking datasets for model efficiency, efficient processing of videos in computer vision, Federated learning, distributed learning, and submission instructions. All papers must be submitted through the Conference Management Toolkit (CMT) link.   https://cmt3.research.microsoft.com/RCV2023
All papers need to be formatted as per the ICCV 2023 guidelines. Please use the template made available at  iccv2023AuthorGuidelines.
Papers can be submitted in two different formats: 
⦁ Long paper:  Long papers should not be not more than 8 pages in length excluding references. All the requirements are the same as the ICCV 2023 paper style guide.
⦁ Short paper: Short papers should not exceed 4 pages excluding references. All other requirements are the same as the ICCV 2023 paper style guide. 

Note that the submission for both types of papers is using the same portal. Papers less than 4 pages will be reviewed as short papers, those more than 4 pages  (excluding references) will be reviewed as long papers, and papers beyond 8 pages in length (excluding references) will be desk rejected. All the accepted papers will be included as part of the ICCV workshop proceedings.

------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Workshop Challenges
Timeline of the challenges: 
⦁ Competitions Start: April 12, 2023
⦁ Competitions End: July 15, 2023: 
⦁ Submission deadline for code and report: July 25, 2023
⦁ Notification of results: Aug 05, 2023
FINAL LEADERBOARD (updated 11 Apr, 2023)

---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Track I: Budgeted Model Training challenge
The aim of this competition track is to train a deep learning model for the task of classification with a constraint of 6 GB GPU memory limit. This implies that the maximum GPU memory usage at any point in the entire training process should not exceed 6 GB. Any solution that violates this constraint will not qualify. The training time is limited to 9 GPU hours  [START THE CHALLENGE HERE]

Track II: Budgeted Model Inference challenge
This competition requires that the resultant model is light-weight and the latency measure is below X on a YY GPU or similar. Solutions will be ranked in terms of accuracy, however, all solutions that violate the latency budget during inference will be rejected. [START THE CHALLENGE HERE]

--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
PRIZES:
1. Details related to prizes will follow soon.
2. We will help two participants in total (from developing countries) through partial financial support to attend ICCV 2023 and present their works. More details to be updated soon.
RULES
1. Both the competitions will be hosted as Kaggle community competitions.
2. Each challenge will be organized in two phases. Phase I corresponds to the standard classification problem and the Kaggle leaderboard of the respective competitions will be used to rank the solutions. Phase II of the competitions will be using a different evaluation metric that also considers the training time/inference time. 
3. Only those solutions of Phase I qualify for Phase II that rank above our baseline solutions on the Kaggle leaderboard.
4. Phase II will commence on May 05, 2023. All qualifying participants would be required to submit the code for their solution.
5. From May 05 onwards, we will run each submitted code with the modified evaluation metrics. For Track I, the training time will be limited to 9 GPU hours of V100 and a maximum memory limit of 6 GB on the GPU. For Track II, the inference time cutoff is XX. More details can be found on the respective competitions page.
6. The top 10 solutions on the final leaderboard would be required to submit their final code and a report of 2 pages describing their solution.
Workshop publication on the challenges
Based on the results and discussions around the two challenges, we will compile challenge papers that will be either submitted for inclusion in ICCV 2023 proceedings or will be submitted as a technical article to a good journal such as TMLR or IJCV.

---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
INVITED SPEAKERS
⦁ Prof. Eric Xing (Petuum and President, MBZUAI; CMU)
⦁ Prof. Song Han (MIT)
⦁ Prof. Anima Anandkumar (Caltech and NVIDIA)
⦁ Prof. Nicholas Lane (Univ. Cambridge and Samsung AI)
⦁ Dr. Prateek Jain (Sr. Staff Research Scientist, Google AI; Adjunct Faculty, IIT Kanpur)
⦁ Dr. Rajat Thomas (AI Engineering Manager, Serket Tech; Weill-Cornell, Qatar)
⦁ Dr. Sangdoo Yun (Research Scientist at Naver AI Research)

----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Main Organizing Committee
1. Dr. Deepak K. Gupta (Assoc. Prof. II, UiT The Arctic University of Norway)
2. Arnav Chavan (Graduate student, IIT Dhanbad; RA, MBZUAI; co-founder, Nyun AI)
3. Rishabh Tiwari (Researcher, Google Research)
4. Gowreesh Mago (Graduate student, IIT Dhanbad)
5. Animesh Gupta (UG student, Thapar University)
6. Dr. Amirhossein Habibian (Research Scientist, Qualcomm AI Research)
7. Dr. Amir Ghodrati (Research Scientist, Qualcomm AI Research)
8. Dr. Babak Ehteshami Bejnordi (Research Scientist, Qualcomm AI Research)
9. Dr. Dilip Prasad (Associate Prof., UiT The Arctic University of Norway; VP, Nyun AI)
10. Dr. Samir Malakar (Postdoc UiT The Arctic University of Norway)
11. Jai Gupta (Senior Software Engineer, Google Research)
12. Dr. Devanshu Arya (Research Scientist, Serket BV)
13. Dr. Sadaf Gulshad (University of Amsterdam)
14. Dr. Zhuang Liu (Research Scientist, FAIR)
15. Dr. Jiahui Yu (Research Scientist, Google Brain)
16. Dr. Zhiqiang Shen (Assistant Prof., MBZUAI)

For any questions, please reach out to guptade...@gmail.com.

best regards,

Deepak Gupta