Research map / Video / Action recognition
Learning From Temporal Gradient for Semi-Supervised Action Recognition
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 cross-modal dense alignment to FixMatch (ablation) | FixMatch · UCF-101 20% | Top-1 accuracy: 54.1 -> 74.6 | Helped |
| Add cross-modal contrastive learning to FixMatch (ablation) | FixMatch · UCF-101 20% | Top-1 accuracy: 54.1 -> 68.6 | Helped |
| Add both dense alignment and cross-modal contrastive learning (ablation) | FixMatch · UCF-101 20% | Top-1 accuracy: 54.1 -> 76.1 | Helped |
| Use fast rather than slow temporal gradient as the teacher (ablation) | RGB student · UCF-101 split-1 20% | Top-1 accuracy: 68.2 -> 74.1 | Helped |
| Use temporal gradient rather than RGB as the input modality (ablation) | FixMatch · UCF-101 split-1 20% | Top-1 accuracy: 52.9 -> 67.3 (slow TG); 52.9 -> 68.3 (fast TG) | Helped |
| Use cosine rather than L1 or L2 dense alignment (ablation) | RGB student with TG teacher · UCF-101 split-1 20% | Top-1 accuracy: 74.0 (L1); 74.4 (L2); 74.6 (Cosine) | Helped |
| Add stop-gradient on the TG teacher side (ablation) | Dense alignment distillation · UCF-101 split-1 20% | Top-1 accuracy: 60.0 -> 74.6 | Helped |
| Average RGB and TG predictions to generate pseudo-labels (ablation) | Two-modality FixMatch · UCF-101 split-1 20% | Top-1 accuracy: 73.6 (RGB); 74.1 (TG); 72.8 (Self); 74.6 (Average) | Helped |
| Align additional earlier feature blocks (ablation) | RGB student with TG teacher · UCF-101 split-1 20% | Top-1 accuracy: 71.4 -> 74.0 -> 74.4 -> 74.6 | Helped |
| Add supervised warm-up before semi-supervised training (ablation) | Proposed method · UCF-101 split-1 20% | Top-1 accuracy: 71.9 -> 74.1 | Helped |
| Add learning-rate warm-up (ablation) | Proposed method · UCF-101 split-1 20% | Top-1 accuracy: 71.1 -> 71.9 | Helped |
| Add PreciseBN (ablation) | Proposed method · UCF-101 split-1 20% | Top-1 accuracy: 74.1 -> 74.6 | Helped |
Closest papers by meaning
- Semi-Supervised Action Recognition With Temporal Contrastive Learning (CVPR 2021 · similarity 0.92)
- TimeBalance: Temporally-Invariant and Temporally-Distinctive Video Representations for Semi-Supervised Action Recognition (CVPR 2023 · similarity 0.92)
- Iterative Contrast-Classify for Semi-supervised Temporal Action Segmentation (AAAI 2022 · similarity 0.90)
- SVFormer: Semi-Supervised Video Transformer for Action Recognition (CVPR 2023 · similarity 0.90)
- SeFAR: Semi-supervised Fine-grained Action Recognition with Temporal Perturbation and Learning Stabilization (AAAI 2025 · similarity 0.88)
- Semi-supervised Active Learning for Video Action Detection (AAAI 2024 · similarity 0.87)
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.