Research map / Video / Action recognition
AViD Dataset: Anonymized Videos from Diverse Countries
AJ Piergiovanni, Michael Ryoo
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 |
|---|---|---|---|
| Train the 3D ResNet-50 on geographically diverse AViD videos rather than Kinetics-400 | 3D ResNet-50 · AViD shared classes, Rest of World | accuracy: 64.5% -> 73.5% | Improved |
| Train on AViD rather than Kinetics-600 | 3D ResNet-50 · AViD shared classes, Rest of World | accuracy: 65.5% -> 73.5% | Improved |
| Train on AViD rather than Kinetics-400 | 3D ResNet-50 · AViD shared classes, North America | accuracy: 72.8% -> 75.2% | Improved |
| Train on AViD rather than Kinetics-400 | 3D ResNet-50 · AViD shared classes, Africa | accuracy: 58.4% -> 71.4% | Improved |
| Pretrain on AViD rather than Kinetics-400 | I3D · HMDB-51 | accuracy: 72.5 -> 75.2 | Improved |
| Pretrain on AViD rather than Kinetics-600 | 3D ResNet-50 · HMDB-51 | accuracy: 76.2 -> 77.3 | Improved |
| Add the AViD no-action examples to pretraining (ablation) | I3D · Charades localization | mAP: 17.3 -> 19.7 | Improved |
| Add the AViD no-action examples to pretraining (ablation) | 3D ResNet-50 · Charades localization | mAP: 18.7 -> 23.2 | Improved |
| Use weak agglomerative-clustering tags instead of full action labels for pretraining (ablation) | 3D ResNet-50 · HMDB-51 | accuracy: 77.3 -> 76.4 | Hurt |
| Use weak affinity-propagation tags instead of full action labels for pretraining (ablation) | 3D ResNet-50 · HMDB-51 | accuracy: 77.3 -> 75.3 | Hurt |
| Blur faces in the AViD training videos (ablation) | 3D ResNet-50 · AViD | accuracy: 48.2 -> 47.9 | Hurt |
| Use more ordered frames rather than a single frame (ablation) | 3D ResNet-50 · AViD | accuracy: 32.5 -> 47.9 | Improved |
Closest papers by meaning
- Learning Human Action Recognition Representations Without Real Humans (NeurIPS 2023 · similarity 0.86)
- Privacy-Preserving Deep Action Recognition: An Adversarial Learning Framework and A New Dataset (TPAMI 2020 · similarity 0.86)
- Privacy Beyond Pixels: Latent Anonymization for Privacy-Preserving Video Understanding (ICLR 2026 · similarity 0.86)
- Learning to Anonymize Faces for Privacy Preserving Action Detection (ECCV 2018 · similarity 0.85)
- Hollywood in Homes: Crowdsourcing Data Collection for Activity Understanding (ECCV 2016 · similarity 0.85)
- Temporal Alignment-Free Video Matching for Few-shot Action Recognition (CVPR 2025 · similarity 0.83)
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.
