Research map / Label-efficient and robust learning / Semi-supervised learning
Cross-Model Pseudo-Labeling for Semi-Supervised Action Recognition
Yinghao Xu, Fangyun Wei, Xiao Sun, Ceyuan Yang, Yujun Shen, Bo Dai, Bolei Zhou, Stephen Lin
Check 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 |
|---|---|---|---|
| Use cross-model pseudo-labeling instead of FixMatch self-labeling | 3D-ResNet50 CMPL · Kinetics-400, 1% labeled | Top-1 accuracy: 8.6 -> 17.6 | Improved |
| Use cross-model pseudo-labeling instead of FixMatch self-labeling | 3D-ResNet50 CMPL · UCF-101, 1% labeled | Top-1 accuracy: 14.8 -> 25.1 | Improved |
| Add auxiliary cross-model supervision to FixMatch (ablation) | 3D-ResNet50 on Kinetics-400 · Kinetics-400, 1% labeled, 50 epochs | Top-1 accuracy: 6.78 -> 12.04 | Improved |
| Add different frame rates and temporal locations (ablation) | CMPL with auxiliary supervision · Kinetics-400, 1% labeled, 50 epochs | Top-1 accuracy: 12.04 -> 13.71 | Improved |
| Use a different-architecture auxiliary network rather than the identical primary architecture (ablation) | CMPL · Kinetics-400, 1% labeled, 50 epochs | Top-1 accuracy: 6.78 -> 9.97 | Improved |
| Use 3D-ResNet50 × 1/4 as the auxiliary network instead of 3D-ResNet50 × 1/2 (ablation) | CMPL · Kinetics-400, 1% labeled, 50 epochs | Top-1 accuracy: 11.57 -> 12.90 | Improved |
| Increase auxiliary sampling from 8 × 8 to 16 × 4 (ablation) | CMPL · Kinetics-400, 1% labeled, 50 epochs | Top-1 accuracy: 12.04 -> 12.90 | Improved |
| Increase auxiliary sampling from 16 × 4 to 32 × 2 (ablation) | CMPL · Kinetics-400, 1% labeled, 50 epochs | Top-1 accuracy: 12.90 -> 12.94 | Improved |
| Increase temporal offset from t_s 1 to t_s 2 (ablation) | CMPL · Kinetics-400, 1% labeled, 50 epochs | Top-1 accuracy: 12.90 -> 12.64 | Hurt |
| Increase temporal offset from t_s 2 to t_s 10 (ablation) | CMPL · Kinetics-400, 1% labeled, 50 epochs | Top-1 accuracy: 12.64 -> 13.71 | Improved |
| Add both supervised and unsupervised losses to the auxiliary network (ablation) | CMPL · Kinetics-400, 1% labeled, 50 epochs | Top-1 accuracy: 6.78 -> 12.90 | Improved |
| Average the two networks' predictions rather than use cross-model pseudo-labeling (ablation) | CMPL pseudo-labeling · Kinetics-400, 1% labeled, 50 epochs | Top-1 accuracy: 12.90 -> 12.16 | Hurt |
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.
- FixMatch — FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence (NeurIPS 2020) · better on 9 of 9 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
- FinePseudo: Improving Pseudo-Labelling through Temporal-Alignablity for Semi-Supervised Fine-Grained Action Recognition (ECCV 2024 · similarity 0.89)
- Learning from Weakly-Labeled Web Videos via Exploring Sub-concepts (AAAI 2022 · similarity 0.88)
- Rethinking Pseudo-Label Guided Learning for Weakly Supervised Temporal Action Localization from the Perspective of Noise Correction (AAAI 2025 · similarity 0.87)
- SeFAR: Semi-supervised Fine-grained Action Recognition with Temporal Perturbation and Learning Stabilization (AAAI 2025 · similarity 0.87)
- CGMatch: A Different Perspective of Semi-supervised Learning (CVPR 2025 · similarity 0.86)
- Non-Stationary Predictions May Be More Informative: Exploring Pseudo-Labels with a Two-Phase Pattern of Training Dynamics (ICML 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.
