Research map / Video / Action recognition
A Large-Scale Robustness Analysis of Video Action Recognition Models
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 a transformer-based model rather than a CNN | Video action recognition · Kinetics400-P | mean relative robustness γ^r: .80 -> .91 | Helped |
| Use a transformer-based model rather than a CNN | Video action recognition · SSv2-P | mean relative robustness γ^r: .76 -> .77 | Mixed |
| Add Kinetics-400 pretraining (ablation) | Action recognition models · UCF101-P and HMDB51-P | robustness | Helped |
| Add Kinetics-400 pretraining to a transformer rather than a CNN (ablation) | Action recognition models · UCF101-P and HMDB51-P | robustness | Helped |
| Train with spatial perturbations rather than clean data (ablation) | CNN-based models · UCF101-P | robustness | Helped |
| Train with a mix of perturbations (ablation) | ResNet50 · UCF101-P | robustness | Hurt |
| Train with perturbed data (ablation) | MViT · UCF101-DS | relative robustness | Hurt |
| Train with perturbed data (ablation) | ResNet50 · UCF101-DS | relative robustness | Helped |
| Increase temporal perturbation severity | Action recognition models · Kinetics400-P | relative robustness γ^r: .97 | No clear effect |
| Apply temporal perturbations | Action recognition models · SSv2-P | relative robustness γ^r: .50 | Hurt |
| Use a higher-capacity model | Action recognition models · UCF101-P and HMDB51-P | robustness | No clear effect |
Closest papers by meaning
- Multi-dataset Training of Transformers for Robust Action Recognition (NeurIPS 2022 · similarity 0.89)
- Benchmarking the Robustness of Temporal Action Detection Models Against Temporal Corruptions (CVPR 2024 · similarity 0.89)
- DirecFormer: A Directed Attention in Transformer Approach to Robust Action Recognition (CVPR 2022 · similarity 0.88)
- Revealing the unseen: Benchmarking video action recognition under occlusion (NeurIPS 2023 · similarity 0.88)
- DVANet: Disentangling View and Action Features for Multi-View Action Recognition (AAAI 2024 · similarity 0.85)
- TrAction: Action Recognition with Sparse Trajectories (NeurIPS 2026 · 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.