Research map / Video / Video-language understanding
Disentangling Spatial and Temporal Learning for Efficient Image-to-Video Transfer Learning
Zhiwu Qing, Shiwei Zhang, Ziyuan Huang, Yingya Zhang, Changxin Gao, Deli Zhao, Nong Sang
Read the paperCheck your idea against 137,217 papers
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add the temporal encoder to the spatial encoder alone (ablation) | DiST · SSV2, ViT-B/16 8 frames | Top-1 accuracy: 55.0 -> 63.2 | Improved |
| add integration branch to the spatial encoder alone (ablation) | DiST · SSV2, ViT-B/16 8 frames | Top-1 accuracy: 55.0 -> 65.9 | Improved |
| add the temporal encoder on top of the integration branch (ablation) | DiST · SSV2, ViT-B/16 | Top-1 accuracy: 65.9 -> 68.7 | Improved |
| add bidirectional feature interactions between temporal encoder and integration branch (ablation) | DiST · SSV2, ViT-B/16 | Top-1 accuracy: 67.5 -> 68.7 | Improved |
| use R(2+1)D as T-Block rather than TAdaConv, C3D or joint transformer (ablation) | DiST temporal encoder · SSV2 | Top-1 accuracy: 67.8 -> 68.7 | Improved |
| increase temporal encoder frames gamma from 1 to 2 (ablation) | DiST · SSV2 | Top-1 accuracy: 67.9 -> 68.7 | Improved |
| increase temporal encoder frames beyond gamma=4 (ablation) | DiST · SSV2 | Top-1 accuracy: 69.1 (gamma=4) -> 68.5 (gamma=8) | No clear effect |
| widen integration branch from alpha=1/8 to 1/2 (ablation) | DiST · SSV2 | Top-1 accuracy: 62.6 -> 68.7 | Improved |
| widen integration branch from alpha=1/2 to 1 (ablation) | DiST · SSV2 | Top-1 accuracy: 68.7 -> 68.7 | No clear effect |
| use nearest interpolation instead of deconvolution for upsampling (ablation) | DiST feature interaction · SSV2 | Top-1 accuracy: 67.8 -> 68.7 | Improved |
| use low-level features from all ViT layers rather than only deep layers 9-12 (ablation) | DiST · SSV2 | Top-1 accuracy: 66.0 -> 68.7 | Improved |
| use DiST disentangled design instead of EVL decoder with frozen CLIP | frozen CLIP ViT-B/16 · SSV2 32 frames | Top-1 accuracy: 62.4 -> 70.9 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ST-Adapter — ST-Adapter: Parameter-Efficient Image-to-Video Transfer Learning (NeurIPS 2022) · better on 6 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Emergent Temporal Correspondences from Video Diffusion Transformers (NeurIPS 2025 · similarity 0.87)
- VideoRefer Suite: Advancing Spatial-Temporal Object Understanding with Video LLM (CVPR 2025 · similarity 0.87)
- TTVidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining (NeurIPS 2026 · similarity 0.87)
- DFDNet: Disentangling and Filtering Dynamics for Enhanced Video Prediction (AAAI 2025 · similarity 0.86)
- DisMo: Disentangled Motion Representations for Open-World Motion Transfer (NeurIPS 2025 · similarity 0.86)
- Efficient Transfer Learning for Video-language Foundation Models (CVPR 2025 · similarity 0.86)
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.
