Research map / Video / Action recognition
Why Can't I Dance in the Mall? Learning to Mitigate Scene Bias in Action Recognition
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 |
|---|---|---|---|
| pre-train with scene adversarial + human mask confusion losses | 3D-ResNet-18 · HMDB-51 | top-1 accuracy: 53.6 -> 56.7 | Helped |
| pre-train with debiasing losses | 3D-ResNet-18 · UCF-101 | top-1 accuracy: 83.5 -> 84.5 | Helped |
| pre-train with debiasing losses | 3D-ResNet-18 · Diving48 | top-1 accuracy: 18.0 -> 20.5 | Helped |
| pre-train with debiasing losses | R-C3D (3D-ResNet-18) · THUMOS-14 | avg video mAP: 36.7 -> 38.2 | Helped |
| pre-train with debiasing losses | ROAD (VGG-16) · JHMDB | frame mAP@0.5: 32.5 -> 34.5 | Helped |
| use soft instead of hard pseudo scene labels (ablation) | scene adversarial debiasing · HMDB-51 | top-1 accuracy: 54.5 -> 56.2 | Helped |
| add scene adversarial loss with hard pseudo labels (ablation) | 3D-ResNet-18 · HMDB-51 | top-1 accuracy: 53.6 -> 54.5 | Helped |
| add scene adversarial loss (ablation) | model with L_Ent · HMDB-51 | top-1 accuracy: 55.1 -> 56.7 | Helped |
| add human mask confusion loss (ablation) | model with L_Adv · HMDB-51 | top-1 accuracy: 56.2 -> 56.7 | Helped |
| add debiasing losses | 3D-ResNet-18 · Mini-Kinetics-200 val | scene classification accuracy: 29.7 -> 2.9 | Helped |
Closest papers by meaning
- Enhancing Unsupervised Video Representation Learning by Decoupling the Scene and the Motion (AAAI 2021 · similarity 0.90)
- Enabling Detailed Action Recognition Evaluation Through Video Dataset Augmentation (NeurIPS 2022 · similarity 0.88)
- ALBAR: Adversarial Learning approach to mitigate Biases in Action Recognition (ICLR 2025 · similarity 0.87)
- Video Representation Learning Using Discriminative Pooling (CVPR 2018 · similarity 0.87)
- Disentangled Concepts Speak Louder Than Words: Explainable Video Action Recognition (NeurIPS 2025 · similarity 0.86)
- DVANet: Disentangling View and Action Features for Multi-View Action Recognition (AAAI 2024 · similarity 0.85)
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.