The authors first fine-tune a pretrained text-to-video diffusion model with bidirectional attention to act as a powerful offline teacher for object erasing.
StreamErase enables real-time streaming video object removal at 26 FPS
The method transfers erasing capability from a bidirectional diffusion model to a causal autoregressive model, enabling interactive editing with constant latency.
Chinese Tech
Zhuobai Dong · Xiaobing Tu · Xiantao Zhang · Jinkui Ren · Yinggui Wang · Yue Ma · +1 more
Wuhan University · Alibaba Group · Hong Kong University of Science and Technology · Shanghai Jiao Tong University
Research Digest··2 min read
06 FPS on a single H200 GPU.
Why this paper
From Alibaba Group and 3 others
In one line
StreamErase enables sub-second, real-time streaming video object removal at over 26 frames per second on a single GPU.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§