Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Going Deeper into Recognizing Actions in Dark Environments: A Comprehensive Benchmark Study
Yuecong Xu, Haozhi Cao, Jianxiong Yin, Zhenghua Chen, Xiaoli Li, Zhengguo G. Li, Qianwen Xu, Jianfei Yang
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 GMFlow instead of TV-L1 optical flow | SlowOnly · ARID-plus UG2-2.1 | top-1 accuracy: 57.25% -> 59.74% | Improved |
| use optical flow input instead of RGB | SlowOnly · ARID-plus UG2-2.1 | top-1 accuracy: 27.08% -> 57.25% | Improved |
| add RGB stream to flow (two-stream fusion) with TV-L1 | SlowOnly · ARID-plus UG2-2.1 | top-1 accuracy: 57.25% -> 50.61% | Hurt |
| apply off-the-shelf low-light enhancement (Zero-DCE) to dark videos (ablation) | six AR models · ARID-plus UG2-2.1 | top-1 accuracy: average improvement of 5.57% | Mixed |
| apply any off-the-shelf enhancement to 3D-ResNet-50 (ablation) | 3D-ResNet-50 · ARID-plus UG2-2.1 | top-1 accuracy: 17.02% average accuracy gain | Improved |
| apply any off-the-shelf enhancement to TSM (ablation) | TSM · ARID-plus UG2-2.1 | top-1 accuracy: average loss of 7.65% | Hurt |
| use newer X3D-M instead of 3D-ResNeXt-101 | AR backbone · ARID-plus UG2-2.1 (RGB) | top-1 accuracy: 34.45% -> 20.25% | Hurt |
| add DANN domain adaptation vs source-only | 3D-ResNeXt-101 · ARID-plus UG2-2.2 | top-1 accuracy: 27.42% -> 44.04% | Improved |
| add MCD domain adaptation vs source-only | I3D · ARID-plus UG2-2.2 | top-1 accuracy: 26.38% -> 27.15% | No clear effect |
| add MK-MMD domain adaptation vs source-only | 3D-ResNet-50 · ARID-plus UG2-2.2 | top-1 accuracy: 24.10% -> 26.90% | Improved |
| add Zero-DCE enhancement (ablation) | R(2+1)D-Bert (Artificially Inspired) · UG2-2.1 test | top-1 accuracy: +2.98% | Improved |
| add optical flow as extra input in challenge solutions | challenge teams · UG2-2.1 leaderboard | top-1 accuracy: surpassed by most pure RGB teams by more than 2.5% | No clear effect |
Closest papers by meaning
- DarkAct: A RGB-Thermal Dataset and Fusion Framework for Multimodal Low-Light Action Recognition (CVPR 2026 · similarity 0.88)
- AVA: A Video Dataset of Spatio-Temporally Localized Atomic Visual Actions (CVPR 2018 · similarity 0.87)
- DarkShake-DVS: Event-based Human Action Recognition under Low-light and Shaking Camera Conditions (CVPR 2026 · similarity 0.86)
- VA-AR: Learning Velocity-Aware Action Representations with Mixture of Window Attention (AAAI 2025 · similarity 0.84)
- TrAction: Action Recognition with Sparse Trajectories (NeurIPS 2026 · similarity 0.84)
- HARDVS: Revisiting Human Activity Recognition with Dynamic Vision Sensors (AAAI 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.
