Thank @ak92501 for tweeting our latest work.
TL;DR: We find supervised classification objective is compatible with self-supervised masked auto-encoders.
Codes are available. Feel free to check it out!
SupMAE: Supervised Masked Autoencoders Are Efficient Vision Learners
abs: arxiv.org/abs/2205.14540
github: github.com/cmu-enyac/supm…
SupMAE is efficient and can achieve comparable performance with MAE using only 30% compute when evaluated on ImageNet with the ViT-B/16 model
Thanks @_akhaliq for featuring our work! We propose a new way to use imperfect optical flow for vid2vid synthesis. It's been an amazing experience working with @BiWuen, Jialiang, Licheng, Kunpeng, Yinan, @imisra_ , @jbhuang0604 , Peizhao, Peter and @dianamarculescu .
Meta just announced FlowVid
Taming Imperfect Optical Flows for Consistent Video-to-Video Synthesis
paper page: huggingface.co/papers/2312.17…
Diffusion models have transformed the image-to-image (I2I) synthesis and are now permeating into videos. However, the advancement of
Thank you, @_akhaliq, for sharing our work. We're honored to have your support.
StreamV2V introduces a feature bank that links current and previous data to improve temporal consistency while maintaining high efficiency.
Also, a big shout out to @cumulo_autumn and @Chenfeng_X!
Looking Backward: Streaming Video-to-Video Translation with Feature Banks
This paper introduces StreamV2V, a diffusion model that achieves real-time streaming video-to-video (V2V) translation with user prompts. Unlike prior V2V methods using batches to process limited frames,
Thank @arankomatsuzaki for twittering our work.
TL, DR: StreamV2V proposes a backward-looking principle that relates the present to the past to enhance temporal consistency while keeping high efficiency.
Shout out to my co-workers @cumulo_autumn@Chenfeng_X
Looking Backward: Streaming Video-to-Video Translation with Feature Banks
Runs 20 FPS on one A100 GPU, being 15×, 46×, 108×, and 158× faster than FlowVid, CoDeF, Rerender, and TokenFlow, respectively
proj: jeff-liangf.github.io/projects/strea…
abs: arxiv.org/abs/2405.15757
🎥 Today we’re premiering Meta Movie Gen: the most advanced media foundation models to-date.
Developed by AI research teams at Meta, Movie Gen delivers state-of-the-art results across a range of capabilities. We’re excited for the potential of this line of research to usher in
