The Second Workshop on Efficient and On-Device Generation (EDGE) at CVPR 2025 will focus on the latest advancements of generative AI in the computer vision domain, with an emphasis on efficiencies across multiple aspects. We encourage techniques that enable generative models to be trained more efficiently and run on resource-constrained devices, such as mobile phones and edge devices. Through these efforts, we envision a future where these permeating generative AI capabilities become significantly more accessible with virtuous scalability and plateauing carbon footprint.

The topics involved in the workshop include but are not limited to:

• Modeling efficiency:
• Design of efficient modules and algorithms for generative models.
• Efficient architecture for generative models: auto-encoders, diffusion, GAN, autoregressive models, etc.
• Efficient transformers and state-space models.
• Training efficiency:
• Methods for reducing the memory footprint of training generative models.
• Using knowledge distillation to support model compression.
• Hardware-aware training acceleration and energy efficient computing.
• Online training under edge constraints.
• Parameter efficient tuning.
• Mixed pre-training to enable downstream tasks via lightweight tuning.
• Inference efficiency:
• Techniques for reducing the computational complexity of generative models, such as quantization, sparsification, and MoEs.
• Advanced sampling techniques for fast inference.
• Approaches for efficient deployment and distribution generative models across multiple devices or edge devices.
• Hardware-aware inference acceleration and energy efficient computing.
• Data efficiency:
• Few-shot diffusion model.
• Generative model finetuning.
• Training generative models with fewer data.
• Application-specific efficiency:
• Applications of efficient generative AI in computer vision, such as image / video generation, image / video editing, style transfer, and super-resolution.
• Efficient personalization and post-generation editing.
• Efficient generative models for real-time video synthesis: talking head, vitual try-on, gaming engine, etc.

Format: Submissions must use the CVPR 2025 Author Kit for Latex/Word Zip file and follow the CVPR 2025 author instructions and submission policies. Submissions need to be anonymized. A paper accepted to CVPR 2025 main conference can be resubmitted to our Extended Abstract track. A paper accepted by another venue can be resubmitted to our Extended Abstract track if allowed by that venue. Any submission to another CVPR 2025 Workshop cannot be resubmitted to EDGE 2025. The workshop considers two submission tracks:

• Long Paper: submissions are limited to 8 pages excluding references, no appendix or supplementary;
• Extended Abstract: submissions are limited to 4 pages excluding references, no appendix or supplementary.

Only long papers will be included in the CVPR 2025 proceedings.

===

Submission Site: https://openreview.net/group?id=thecvf.com/CVPR/2025/Workshop/EDGE

Submission Deadline: March 21, 2025 (AOE)

Workshop Date: June 12, 2025, 13:00-17:00

Workshop Location: Room: 208 A

Poster Session: June 12, 2025, 12:00-13:00 (before the workshop starts)

Poster Session Location: ExHall D, Board #202-214

We are pleased to announce the accepted papers for the Second Workshop on Efficient and On-Device Generation (EDGE) at CVPR 2025. Congratulations to all authors!

Long Papers

• Title: Geometric Consistency Refinement for Single Image Novel View Synthesis via Test-Time Adaptation of Diffusion Models
  Authors: Josef Bengtson (Chalmers University of Technology), David Nilsson (Chalmers University of Technology), Fredrik Kahl (Chalmers University of Technology)
  [Poster # 202]
• Title: AdaVid: Adaptive Video-Language Pretraining
  Authors: Chaitanya Patel (Stanford University), Juan Carlos Niebles (Salesforce Research), Ehsan Adeli (Stanford University)
  [Poster # 203]
• Title: LLMPi: Optimizing LLMs for High-Throughput on Raspberry Pi
  Authors: Mahsa Ardakani (University of South Carolina), Jinendra Malekar (University of South Carolina), Ramtin Zand (University of South Carolina)
  [Poster # 204]
• Title: Latent Patched Efficient Diffusion Model For High Resolution Image Synthesis
  Authors: Weiyun Jiang (Rice University), Devendra Kumar Jangid (Samsung Research America), Seok-Jun Lee (Samsung Research America), Hamid Rahim Sheikh (Samsung Research America)
  [Poster # 205]
• Title: ADAPTOR: Adaptive Token Reduction for Video Diffusion Transformers
  Authors: Elia Peruzzo (University of Trento), Adil Karjauv (Qualcomm), Nicu Sebe (University of Trento), Amir Ghodrati (Qualcomm AI Research), Amir Habibian (Qualcomm)
  [Poster # 206]
• Title: Scaling On-Device GPU Inference for Large Generative Models
  Authors: Jiuqiang Tang (Google), Raman Sarokin (Google), Ekaterina Ignasheva (Meta), Grant Jensen (Google), Lin Chen (Google), Juhyun Lee (Google), Andrei Kulik (Google), Matthias Grundmann (Google)
  [Poster # 207]

Short Papers

• Title: Random Conditioning for Diffusion Model Compression with Distillation
  Authors: Dohyun Kim (Korea University), Sehwan Park (Korea University), Geonhee Han (Korea University), Seung Wook Kim (Nvidia), Paul Hongsuck Seo (Korea University)
  [Poster # 208]
• Title: LatentLLM: Attention-Aware Joint Tensor Compression
  Authors: Toshiaki Koike-Akino (Mitsubishi Electric Research Labs), Xiangyu Chen (Mitsubishi Electric Research Labs), Jing Liu (Mitsubishi Electric Research Labs), Ye Wang (Mitsubishi Electric Research Labs), Pu Perry Wang (Mitsubishi Electric Research Labs), Matthew Brand (Mitsubishi Electric Research Labs)
  [Poster # 209]
• Title: MS-Temba: Multi-Scale Temporal Mamba for Efficient Temporal Action Detection
  Authors: Arkaprava Sinha (University of North Carolina at Charlotte), Monish Soundar Raj (University of North Carolina at Charlotte), Pu Wang (University of North Carolina at Charlotte), Ahmed Helmy (University of North Carolina at Charlotte), Srijan Das (University of North Carolina at Charlotte)
  [Poster # 210]
• Title: TuneComp: Joint Fine-Tuning and Compression for Large Foundation Models
  Authors: Xiangyu Chen (Mitsubishi Electric Research Labs), Jing Liu (Mitsubishi Electric Research Labs), Ye Wang (Mitsubishi Electric Research Labs), Matthew Brand (Mitsubishi Electric Research Labs), Pu Perry Wang (Mitsubishi Electric Research Labs), Toshiaki Koike-Akino (Mitsubishi Electric Research Labs)
  [Poster # 211]
• Title: Efficient Personalization of Quantized Diffusion Model without Backpropagation
  Authors: Hoigi Seo (Seoul National University), Wongi Jeong (Seoul National University), Kyungryeol Lee (Seoul National University), Se Young Chun (Seoul National University)
  [Poster # 212]
• Title: Atlas: Multi-Scale Attention Improves Long Context Image Modeling
  Authors: Kumar Krishna Agrawal (University of California, Berkeley), Long Lian (University of California, Berkeley), Longchao Liu (University of California, Berkeley), Natalia Harguindeguy (University of California, Berkeley), Boyi Li (Nvidia Research), Alexander G Bick (Vanderbilt University), Maggie Chung (University of California, San Francisco), Trevor Darrell (University of California, Berkeley), Adam Yala (University of California, Berkeley)
  [Poster # 213]
• Title: LoRA.rar: Learning to Merge LoRAs via Hypernetworks for Subject-Style Conditioned Image Generation
  Authors: Donald Shenaj (Samsung R&D Institute UK (SRUK), University of Padova), Ondrej Bohdal (Samsung R&D Institute UK (SRUK)), Mete Ozay (Samsung R&D Institute UK (SRUK)), Pietro Zanuttigh (University of Padova), Umberto Michieli (Samsung R&D Institute UK (SRUK))
  [Poster # 214]

We are pleased to announce two paper awards at CVPR 2025 EDGE workshop, which will be sponsored by PixVerse. Congratulations to the award winners!

Best Paper Award

• Title: AdaVid: Adaptive Video-Language Pretraining
  Authors: Chaitanya Patel (Stanford University), Juan Carlos Niebles (Salesforce Research), Ehsan Adeli (Stanford University)
  [Poster # 203]

Best Paper Runner-Up Award

• Title: Scaling On-Device GPU Inference for Large Generative Models
  Authors: Jiuqiang Tang (Google), Raman Sarokin (Google), Ekaterina Ignasheva (Meta), Grant Jensen (Google), Lin Chen (Google), Juhyun Lee (Google), Andrei Kulik (Google), Matthias Grundmann (Google)
  [Poster # 207]