Research map / Video / Video segmentation and tracking
Benchmarking the Robustness of Spatial-Temporal Models Against Corruptions
Chenyu Yi, SIYUAN YANG, Haoliang Li, Yap-peng Tan, Alex Kot
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 |
|---|---|---|---|
| use transformer (TimeSformer) instead of CNN | video classifier · Mini Kinetics-C | mPC: 71.4 | Improved |
| scale backbone from ResNet18 to ResNet50 | 3D ResNet · Mini Kinetics-C | mPC: 53.3 -> 59.2 | Improved |
| scale backbone from ResNet18 to ResNet50 | 3D ResNet · Mini SSV2-C | mPC: 42.6 -> 46.6 | Improved |
| reduce FLOPs from 3D ResNet-50 to SlowFast | 3D ResNet-50 · Mini Kinetics-C | rPC: 81.1 -> 78.5 | Hurt |
| reduce FLOPs from 3D ResNet-50 to X3D-M | 3D ResNet-50 · Mini Kinetics-C | rPC: 81.1 -> 77.6 | Hurt |
| train with Gaussian noise std 0.1 (ablation) | 3D ResNet-18 · Mini Kinetics-C | mPC: 53.3 -> 50.1 | Hurt |
| train with Gaussian noise std 0.2 (ablation) | 3D ResNet-18 · Mini Kinetics-C | mPC: 53.3 -> 43.8 | Hurt |
| train with Gaussian noise std 0.1 (ablation) | 3D ResNet-18 · Mini SSV2-C | mPC: 42.6 -> 36.0 | Hurt |
| train with Gaussian noise std 0.2 (ablation) | 3D ResNet-18 · Mini SSV2-C | mPC: 42.6 -> 36.2 | Hurt |
| train with Gaussian noise std 0.1 (ablation) | 3D ResNet-18 · Mini Kinetics-C shot noise | accuracy: 47.7 -> 61.9 | Improved |
| train with Gaussian noise std 0.1 (ablation) | 3D ResNet-18 · Mini Kinetics-C fog | accuracy: 40.5 -> 28.7 | Hurt |
| train on single corruption (Fog) (ablation) | 3D ResNet-18 · 40-class Mini Kinetics-C | mPC: 65.8 -> 41.8 | Hurt |
Closest papers by meaning
- Benchmarking the Robustness of Temporal Action Detection Models Against Temporal Corruptions (CVPR 2024 · similarity 0.87)
- Instant Video Models: Universal Adapters for Stabilizing Image-Based Networks (NeurIPS 2025 · similarity 0.87)
- ImageNet-D: Benchmarking Neural Network Robustness on Diffusion Synthetic Object (CVPR 2024 · similarity 0.86)
- Dynamic Sparse Training versus Dense Training: The Unexpected Winner in Image Corruption Robustness (ICLR 2025 · similarity 0.86)
- LAION-C: An Out-of-Distribution Benchmark for Web-Scale Vision Models (ICML 2025 · similarity 0.85)
- Revealing the Dark Secrets of Extremely Large Kernel ConvNets on Robustness (ICML 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.
