AI Paper Atlas

Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning

Zero-Shot Compositional Video Learning with Coding Rate Reduction

Heeseok Jung, Jun-Hyeon Bak, Yujin Jeong, Gyugeun Lee, Jinwoo Ahn, Eun-Sol Kim

ICCV 2025 · Label-efficient and robust learning › Zero-/few-shot and meta-learning · approach: Generative feature synthesis for zero-shot · 1 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add spatiotemporal position embeddings to resampler (ablation)coding-rate-reduction resampler · Sth-Com (VideoSwin-T)HM: 28.1 -> 36.8Improved
add rate reduction optimization step on top of position embeddings (ablation)coding-rate-reduction resampler · Sth-Com (VideoSwin-T)HM: 36.8 -> 37.3Improved
add rate reduction optimization step on top of position embeddings (ablation)coding-rate-reduction resampler · Sth-Com (VideoSwin-T)AUC: 18.0 -> 18.3Improved
add rate reduction optimization step (ablation)coding-rate-reduction resampler · Sth-Com test splitentanglement coding rate: 508.4 -> 504.3Improved
use too many latent queries (M=256 vs 128) (ablation)resampler · Sth-Com (VideoSwin-T)HM: 37.3 -> 36.6Hurt
set number of subspaces K to 24 (heads) rather than 6, 12 or 48 (ablation)resampler · Sth-Com (VideoSwin-T)HM: 36.5 -> 37.3Improved
replace label-embedding concatenation and auxiliary training of C2C with the rate reduction resamplerSth-Com (VideoSwin-T) · Sth-Com (VideoSwin-T + fastText)HM: 36.7 -> 37.3 vs C2C (Enhanced)Improved
use CLIP+AIM backbone with the proposed resampler instead of C2CSth-Com · Sth-Com (CLIP + AIM)HM: 41.0 -> 42.6Improved
proposed method versus fine-tuned VLMs on CATERCATER · CATER unseen mAPmAP unseen: 37.0 -> 74.0 vs InternVL2-8BImproved
proposed method versus C2C on object accuracy with CLIPSth-Com (CLIP + AIM) · Sth-Com (CLIP + AIM)object accuracy: 56.2 -> 54.7Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy