Research map / Video / Action recognition
Learning from Weakly-Labeled Web Videos via Exploring Sub-concepts
Read the paperCheck 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 SPL-B instead of ImageNet pre-training | R3D-50-G · Kinetics-200 | Top-1 accuracy: 80.6 -> 84.3 | Helped |
| Use SPL-B labels instead of weak labels | R3D-50-G · Kinetics-200 | Top-1 accuracy: 82.8 -> 84.3 | Helped |
| Use SPL labels instead of teacher predictions alone | R3D-50-G · Kinetics-200 | Top-1 accuracy: 81.9 -> 84.3 | Helped |
| Use SPL-B instead of agreement filtering | R3D-50-G · Kinetics-200 | Top-1 accuracy: 82.9 -> 84.3 | Helped |
| Use SPL-B instead of Data Parameters | R3D-50-G · Kinetics-200 | Top-1 accuracy: 83.2 -> 84.3 | Helped |
| Change SPL mapping from binary agreement/disagreement to merging rare off-diagonal classes into diagonals (ablation) | SPL · Kinetics-200 | Top-1 accuracy: 84.9 -> 84.5 | Hurt |
| Change SPL mapping from binary agreement/disagreement to discarding tail off-diagonal samples (ablation) | SPL · Kinetics-200 | Top-1 accuracy: 84.9 -> 84.6 | Hurt |
| Increase the web-video pool from 147K-V to 285K-V at roughly fixed clip count (ablation) | SPL-B · Kinetics-200 | Top-1 accuracy: 84.9 -> 85.3 | Helped |
| Increase the web-video pool from 147K-V to 285K-V at roughly fixed clip count | Weak Label Train · Kinetics-200 | Top-1 accuracy: 83.9 -> 84.0 | No clear effect |
| Use SPL-B pre-training instead of the ImageNet baseline | R3D-50-G · HMDB-51 | accuracy: 46.0 -> 67.6 | Helped |
| Use SPL-B pre-training instead of the ImageNet baseline | R3D-50-G · UCF-101 | accuracy: 84.9 -> 94.6 | Helped |
| Use SPL pre-training instead of the ImageNet baseline | R3D-50-G · SoccerNet | mean average precision: 73.7 -> 76.8 | Helped |
Closest papers by meaning
- Large-Scale Weakly-Supervised Pre-Training for Video Action Recognition (CVPR 2019 · similarity 0.93)
- Semi-Weakly-Supervised Learning of Complex Actions From Instructional Task Videos (CVPR 2022 · similarity 0.89)
- Learning Transferable Self-Attentive Representations for Action Recognition in Untrimmed Videos with Weak Supervision (AAAI 2019 · similarity 0.89)
- Semi-supervised Active Learning for Video Action Detection (AAAI 2024 · similarity 0.87)
- 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 0.86)
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.