🚀 Excited to share our latest work: "EdgeSAM - Prompt-In-the-Loop Distillation for On-Device Deployment of SAM" Supercharged SAM for Edge Devices! 🌟 #EdgeSAM is a faster, optimized version of SAM, now tailored for edge devices. We've reimagined SAM's ViT-based image encoder into a CNN architecture, perfect for these devices. Our unique approach includes distilling the prompt encoder & mask decoder, ensuring our model grasps the complex dynamics of user input & mask generation. 🏎️ EdgeSAM boasts a 40x speed boost over SAM & outperforms MobileSAM, being 14x faster on edge devices. Plus, it enhances mIoUs on COCO & LVIS by 2.3 and 3.2! First SAM variant to run over 30 FPS on iPhone 14. Check out our code, demo & models! 🔗 Project page: 🔗 GitHub: 🤗 Hugging Face: Together with @ChongZhou7, @xtl994 and @doubledaibo

HUGE thank you for your service to our #ICCV2023 outstanding reviewers!

FreeU: Free Lunch in Diffusion U-Net paper page: we uncover the untapped potential of diffusion U-Net, which serves as a "free lunch" that substantially improves the generation quality on the fly. We initially investigate the key contributions of the U-Net architecture to the denoising process and identify that its main backbone primarily contributes to denoising, whereas its skip connections mainly introduce high-frequency features into the decoder module, causing the network to overlook the backbone semantics. Capitalizing on this discovery, we propose a simple yet effective method-termed "FreeU" - that enhances generation quality without additional training or finetuning. Our key insight is to strategically re-weight the contributions sourced from the U-Net's skip connections and backbone feature maps, to leverage the strengths of both components of the U-Net architecture. Promising results on image and video generation tasks demonstrate that our FreeU can be readily integrated to existing diffusion models, e.g., Stable Diffusion, DreamBooth, ModelScope, Rerender and ReVersion, to improve the generation quality with only a few lines of code.

The twelve #CVPR2023 award candidate papers are listed at Congratulations to all of the authors on this achievement!