Research map / Video / Video segmentation and tracking
Day2Dark: Pseudo-Supervised Activity Recognition Beyond Silent Daylight
Yunhua Zhang, Hazel Doughty, Cees G. M. Snoek
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 pseudo-supervised day2dark learning (ablation) | darkness-adaptive recognizer · EPIC-Kitchens (dark) | top-1 accuracy: 29.3 -> 34.2 | Improved |
| add day2dark-mix finetuning (ablation) | darkness-adaptive recognizer · EPIC-Kitchens (dark) | top-1 accuracy: 34.2 -> 35.6 | Improved |
| add pseudo-supervised learning to visual-only model (ablation) | visual encoder · Kinetics-Sound (dark) | top-1 accuracy: 77.2 -> 78.5 | Improved |
| add darkness probe and adaptive encoder (ablation) | vanilla multi-modal transformer · EPIC-Kitchens (dark) | top-1 accuracy: 29.8 -> 31.8 | Improved |
| add adaptive prompts (ablation) | recognizer · EPIC-Kitchens (dark) | top-1 accuracy: 31.8 -> 34.3 | Improved |
| add adaptive classification heads (ablation) | recognizer · EPIC-Kitchens (dark) | top-1 accuracy: 34.3 -> 35.6 | Improved |
| add a second auxiliary task (both video-text matching and sound localization) (ablation) | pseudo-supervised learning · EPIC-Kitchens (dark), auxiliary tasks | top-1 accuracy: 33.3 -> 35.6 | Improved |
| use auxiliary-task predictions of target task as pseudo labels (ablation) | pseudo-supervised learning · EPIC-Kitchens (dark) | top-1 accuracy: 35.6 -> 30.4 | Hurt |
| use raw auxiliary predictions instead of autoencoder bottleneck (ablation) | pseudo-supervised learning · EPIC-Kitchens (dark) | top-1 accuracy: 35.6 -> 31.5 | Hurt |
| replace learned darkness probe with computed illuminance (ablation) | recognizer · EPIC-Kitchens (dark) | top-1 accuracy: -2.1% | Hurt |
| increase lambda from 0.01 to 0.03 (ablation) | pseudo-supervised learning · EPIC-Kitchens (dark), lambda | top-1 accuracy: 35.6 -> 35.2 | Hurt |
| use K=5 branches instead of K=1 (ablation) | recognizer · EPIC-Kitchens (dark), branches K | top-1 accuracy: 29.8 -> 35.6 | Improved |
Closest papers by meaning
- Audio-Adaptive Activity Recognition Across Video Domains (CVPR 2022 · similarity 0.85)
- Going Deeper into Recognizing Actions in Dark Environments: A Comprehensive Benchmark Study (IJCV 2023 · similarity 0.84)
- HARDVS: Revisiting Human Activity Recognition with Dynamic Vision Sensors (AAAI 2024 · similarity 0.83)
- NightCC: Nighttime Color Constancy via Adaptive Channel Masking (CVPR 2024 · similarity 0.82)
- Boosting Object Detection with Zero-Shot Day-Night Domain Adaptation (CVPR 2024 · similarity 0.82)
- TIM: A Time Interval Machine for Audio-Visual Action Recognition (CVPR 2024 · similarity 0.82)
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.
