The authors divide videos into temporal segments and assign each segment learnable abstract tokens that summarize its content.
CoVisco compresses long videos inside the vision encoder
Segmented attention and learned summary tokens let the encoder process long videos while presenting downstream models with a compact, adjustable visual interface.
Chinese Tech
Yulong Liu · Xiaotian Han · Junyuan Shang · Yuchen Ding · Zhenyu Zhang · Shuohuan Wang · +3 more
ERNIE Team, Baidu Inc. · The Hong Kong University of Science and Technology · Institute of Automation, Chinese Academy of Sciences (CASIA)
Research Digest··2 min read
Liu and colleagues introduce CoVisco, a unified image and video encoder designed to avoid both dense attention across every frame and post-hoc token compression.
Why this paper
From Institute of Automation, Chinese Academy of Sciences (CASIA) and 2 others
In one line
CoVisco uses abstract tokens and segmented attention to compress visual tokens, enabling 400-token video understanding that matches or exceeds OneVision-Encoder.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ·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.
§