Research map / Video / Action recognition
Semi-Supervised Action Recognition With Temporal Contrastive Learning
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 |
|---|---|---|---|
| use two-speed temporal contrastive learning instead of FixMatch | ResNet-18 action recognition · Mini-Something-V2, 5% labeled | Top-1 accuracy: 21.67${\pm 0.18}$ -> 29.81${\pm 0.77}$; 8.14% | Helped |
| use two-speed temporal contrastive learning instead of S4L | ResNet-18 action recognition · Jester, 1% labeled | Top-1 accuracy: 64.98$\pm 2.70$ -> 75.21$\pm 4.48$; 10.23% | Helped |
| use two-speed temporal contrastive learning instead of FixMatch | ResNet-18 action recognition · Kinetics-400, 5% labeled | Top-1 accuracy: 25.65$\pm 0.28$ -> 30.28$\pm 0.13$; 4.63% | Helped |
| use TCL with unlabeled first-person and target-domain videos instead of FixMatch | ResNet-18 action recognition · Charades-Ego, 10% labeled, rho=0 | mAP: 17.96$\pm{0.25}$ -> 18.50$\pm{0.95}$ | Helped |
| add pseudo-label finetuning | TCL · Mini-Something-V2, 5% labeled, ResNet-18 | Top-1 accuracy: 29.81${\pm 0.77}$ -> 30.55${\pm 1.36}$ | Helped |
| add self-supervised pretraining and pseudo-label finetuning | TCL · Mini-Something-V2, 5% labeled, ResNet-18 | Top-1 accuracy: 29.81${\pm 0.77}$ -> 30.97${\pm 0.07}$ | Helped |
| add group-contrastive loss (ablation) | TCL · Mini-Something-V2, 5% labeled, ResNet-18 | Top-1 accuracy: 27.24$\pm{0.42}$ -> 29.81$\pm{0.77}$ | Helped |
| replace temporal contrastive loss with pseudo-label consistency loss (ablation) | TCL · Mini-Something-V2, 5% labeled, ResNet-18 | Top-1 accuracy: 29.81$\pm{0.77}$ -> 23.60$\pm{1.04}$; around 6.21% | Hurt |
| increase pathway frame segments from 8/4 to 16/8 (ablation) | TCL · Mini-Something-V2, 5% labeled, ResNet-18 | Top-1 accuracy: 29.81% -> 31.31%; 1.5% | Helped |
| lower instance-contrastive loss weight gamma from 9 to 3 (ablation) | TCL · Mini-Something-V2, 5% labeled, ResNet-18 | Top-1 accuracy: more than 6% | Hurt |
| vary unlabeled-to-labeled ratio mu among 3, 5, and 7 (ablation) | TCL · Mini-Something-V2, 5% labeled, ResNet-18 | Top-1 accuracy | No clear effect |
| replace temporal contrastive learning with O3N (ablation) | TCL · Mini-Something-V2, 5% labeled, ResNet-18 | Top-1 accuracy: 29.81% -> 24.58% | Hurt |
Closest papers by meaning
- TimeBalance: Temporally-Invariant and Temporally-Distinctive Video Representations for Semi-Supervised Action Recognition (CVPR 2023 · similarity 0.93)
- Iterative Contrast-Classify for Semi-supervised Temporal Action Segmentation (AAAI 2022 · similarity 0.92)
- Learning From Temporal Gradient for Semi-Supervised Action Recognition (CVPR 2022 · similarity 0.92)
- RSPNet: Relative Speed Perception for Unsupervised Video Representation Learning (AAAI 2021 · 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.