Research map / Video / Action recognition
FinePseudo: Improving Pseudo-Labelling through Temporal-Alignablity for Semi-Supervised Fine-Grained Action Recognition
Ishan Rajendrakumar Dave, Mamshad Nayeem Rizve, Mubarak Shah
Read the paperCheck 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 |
|---|---|---|---|
| add FinePseudo collaborative pseudo-labeling | R2plus1D-18 · Diving48 (10% labels) | top-1 accuracy: 33.7 -> 37.6 | Improved |
| add FinePseudo to AIM CLIP ViT-B | AIM ViT-B · Diving48 (5% labels) | top-1 accuracy: 38.12 -> 43.02 | Improved |
| add FinePseudo in open-world setting with novel classes | R2plus1D-18 · Diving48 open-world (10% labels) | top-1 accuracy: 39.60 -> 42.21 | Improved |
| use regular pseudo-labeling in open-world setting | R2plus1D-18 · Diving48 open-world (10% labels) | top-1 accuracy: 39.60 -> 38.29 | Hurt |
| add alignability-verification metric learning (both losses) to alignability encoder (ablation) | FinePseudo · Diving48 (10% labels) | top-1 accuracy: 33.50 -> 37.64 | Improved |
| add triplet loss L_AT in addition to score loss (ablation) | FinePseudo · Diving48 (10% labels) | top-1 accuracy: 36.11 -> 37.64 | Improved |
| add GITDL self-supervised pretraining for alignability encoder (ablation) | FinePseudo · Diving48 (10% labels) | top-1 accuracy: 35.23 -> 37.64 | Improved |
| combine alignability encoder with action encoder rather than using either alone (ablation) | FinePseudo · Diving48 (10% labels) | top-1 accuracy: 33.10 -> 37.64 | Improved |
| add uncertainty filtering to confidence-based pseudo-labels (ablation) | pseudo-labeling · Diving48 (10% labels) | top-1 accuracy: 33.40 -> 33.58 | No clear effect |
| verify pseudo-labels with alignability score (ablation) | pseudo-labeling · Diving48 (10% labels) | top-1 accuracy: 33.40 -> 37.09 | Improved |
| use non-parametric alignability classifier for pseudo-labels instead of verification only (ablation) | pseudo-labeling · Diving48 (10% labels) | top-1 accuracy: 37.09 -> 37.64 | Improved |
| use hard negatives instead of all negatives in triplet mining (ablation) | alignability metric learning · Diving48 (10% labels) | top-1 accuracy: 36.16 -> 37.64 | Improved |
Closest papers by meaning
- SeFAR: Semi-supervised Fine-grained Action Recognition with Temporal Perturbation and Learning Stabilization (AAAI 2025 · similarity 0.91)
- Weakly-Supervised Temporal Action Detection for Fine-Grained Videos with Hierarchical Atomic Actions (ECCV 2022 · similarity 0.90)
- Cross-Model Pseudo-Labeling for Semi-Supervised Action Recognition (CVPR 2022 · similarity 0.89)
- Storyboard-guided Alignment for Fine-grained Video Action Recognition (NeurIPS 2025 · similarity 0.89)
- Semi-Supervised Action Recognition With Temporal Contrastive Learning (CVPR 2021 · similarity 0.89)
- FineParser: A Fine-grained Spatio-temporal Action Parser for Human-centric Action Quality Assessment (CVPR 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.
