Research map / Video / Action recognition
Mitigating and Evaluating Static Bias of Action Representations in the Background and the Foreground
Haoxin Li, Yuan Liu, Hanwang Zhang, Boyang Li
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 StillMix augmentation | TSM · Kinetics-400 SCUFO | top-1 accuracy (lower is better): 17.22 -> 5.23 | Improved |
| Add StillMix augmentation | TSM · Kinetics-400 IID | top-1 accuracy: 71.13 -> 71.28 | No clear effect |
| Use ActorCutMix or FAME background replacement | TSM, Swin-T · SCUBA vs SCUFO | SCUBA and SCUFO accuracy: SCUBA 37.39 -> 40.91 (FAME), SCUFO 17.22 -> 18.34 | Mixed |
| Use VideoMix, SDN or BE | TSM, Swin-T · Kinetics-400 SCUBA | top-1 accuracy: 37.39 -> 36.95 (SDN, TSM) | No clear effect |
| Choose biased frames via reference network instead of uniform sampling (ablation) | StillMix · HMDB51 contrasted accuracy, TSM | contrasted accuracy: 31.21 -> 33.14 | Improved |
| Sample low-probability frames from reference network instead of high-probability (ablation) | StillMix · HMDB51 contrasted accuracy, TSM | contrasted accuracy: 33.14 -> 29.17 | Hurt |
| Use reference network of the same architecture as main network (ablation) | StillMix · UCF101 contrasted accuracy, Swin-T | contrasted accuracy: 31.09 -> 32.14 | Improved |
| Mix action labels in StillMix (ablation) | StillMix · UCF101/HMDB51 | performance | Hurt |
| Use StillMix pretraining features for weakly supervised localization | BaSNet on TSM RGB features · THUMOS14 | avg mAP: 0.1810 -> 0.1935 | Improved |
| Add StillMix on conflicting-foreground videos | Swin-T · Kinetics-400 binary conflicting FG | accuracy: 72.83 -> 85.51 | Improved |
Closest papers by meaning
- Quantifying and Learning Static vs. Dynamic Information in Deep Spatiotemporal Networks (TPAMI 2024 · similarity 0.90)
- Why Can't I Dance in the Mall? Learning to Mitigate Scene Bias in Action Recognition (NeurIPS 2019 · similarity 0.88)
- Enabling Detailed Action Recognition Evaluation Through Video Dataset Augmentation (NeurIPS 2022 · similarity 0.88)
- Bias for Action: Video Implicit Neural Representations with Bias Modulation (CVPR 2025 · similarity 0.86)
- ALBAR: Adversarial Learning approach to mitigate Biases in Action Recognition (ICLR 2025 · similarity 0.85)
- FineParser: A Fine-grained Spatio-temporal Action Parser for Human-centric Action Quality Assessment (CVPR 2024 · similarity 0.84)
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.
