Research map / Video / Action recognition
CoLA: Weakly-Supervised Temporal Action Localization With Snippet 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 |
|---|---|---|---|
| Add SniCo contrastive loss to the action-classification loss (ablation) | CoLA on THUMOS'14 · THUMOS'14 test | mAP@0.5: 24.7% -> 32.2% (+7.5%) | Helped |
| Add hard-background refinement (ablation) | CoLA SniCo loss on THUMOS'14 · THUMOS'14 test | mAP@0.5: 29.7% -> 32.2% | Helped |
| Add hard-action refinement (ablation) | CoLA SniCo loss on THUMOS'14 · THUMOS'14 test | mAP@0.5: 30.4% -> 32.2% | Helped |
| Increase the number of negative snippets in SniCo (ablation) | CoLA on THUMOS'14 · THUMOS'14 test | mAP@0.5: 28.9 -> 30.4 -> 31.3 -> 31.9 -> 32.2 | Helped |
| Vary hard-snippet dilation mask M from 4 to 9 with m=3 (ablation) | CoLA hard snippet mining · THUMOS'14 test | mAP@0.5: 30.9, 31.8, 32.2, 32.0, 31.8, 32.1 | Mixed |
| Vary hard-snippet erosion mask m from 0 to 5 with M=6 (ablation) | CoLA hard snippet mining · THUMOS'14 test | mAP@0.5: 30.3, 31.7, 32.0, 32.2, 32.0, 31.9 | Mixed |
| Use CoLA rather than the strongest reported weakly supervised baseline | THUMOS'14 localization · THUMOS'14 test | mAP@0.5: 30.1 -> 32.2 | Helped |
| Use CoLA rather than DGAM | ActivityNet v1.2 localization · ActivityNet v1.2 validation | average mAP: 24.4 -> 26.1 | Helped |
Closest papers by meaning
- Weakly-Supervised Temporal Action Localization by Inferring Salient Snippet-Feature (AAAI 2024 · similarity 0.94)
- Weakly Supervised Temporal Action Localization via Representative Snippet Knowledge Propagation (CVPR 2022 · similarity 0.93)
- Exploring Denoised Cross-Video Contrast for Weakly-Supervised Temporal Action Localization (CVPR 2022 · similarity 0.92)
- Fine-Grained Temporal Contrastive Learning for Weakly-Supervised Temporal Action Localization (CVPR 2022 · similarity 0.92)
- HR-Pro: Point-Supervised Temporal Action Localization via Hierarchical Reliability Propagation (AAAI 2024 · similarity 0.90)
- Bridge the Gap: From Weak to Full Supervision for Temporal Action Localization with PseudoFormer (CVPR 2025 · similarity 0.89)
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.