Research map / Video / Action recognition
Human-Centric Transformer for Domain Adaptive Action Recognition
Kun-Yu Lin, Jiaming Zhou, Wei‐Shi Zheng
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 human encoder to backbone (ablation) | Backbone (DANN) · UCF-HMDB U->H / H->U | ACC (%): 71.1 -> 76.7; 75.1 -> 83.2 | Improved |
| add human encoder to backbone (ablation) | Backbone (DANN) · Kinetics-NecDrone | ACC (%): 22.3 -> 35.1 | Improved |
| add context encoder only to backbone (ablation) | Backbone (DANN) · UCF-HMDB H->U | ACC (%): 75.1 -> 70.9 | Hurt |
| add context encoder only to backbone (ablation) | Backbone (DANN) · UCF-HMDB U->H | ACC (%): 71.1 -> 72.8 | Improved |
| add context encoder and human-context decoder to human encoder (ablation) | Backbone+HmEnc · UCF-HMDB U->H / H->U | ACC (%): 76.7 -> 80.8; 83.2 -> 87.4 | Improved |
| late-fuse human and context encoders instead of the decoder (ablation) | Backbone+HmEnc · UCF-HMDB H->U | ACC (%): 83.2 -> 83.5 | No clear effect |
| add context prototypes (ablation) | HCTransformer · UCF-HMDB U->H / H->U | ACC (%): 78.9 -> 80.8; 85.6 -> 87.4 | Improved |
| add spatial masking module (ablation) | HCTransformer · Kinetics-NecDrone | ACC (%): 35.4 -> 38.6 | Improved |
| add spatial masking module (ablation) | HCTransformer · EPIC-Kitchens-UDA | ACC (%): 43.5 -> 45.3 | Improved |
| replace vanilla TRN with human encoder (ablation) | Backbone · Kinetics-NecDrone | ACC (%): 32.0 -> 35.1 | Improved |
| drop temporal pooling of I3D clip features | HCTransformer (I3D) · UCF-HMDB | ACC (%): +0.8% on UCF->HMDB and +1.1% on HMDB->UCF | Improved |
| use DeepLabV3 instead of FCN human detector (ablation) | HCTransformer · UCF-HMDB U->H / H->U | ACC (%): 80.8 -> 81.4; 87.4 -> 86.9 | Mixed |
Closest papers by meaning
- Shuffle and Attend: Video Domain Adaptation (ECCV 2020 · similarity 0.89)
- Interact Before Align: Leveraging Cross-Modal Knowledge for Domain Adaptive Action Recognition (CVPR 2022 · similarity 0.89)
- Unsupervised Video Domain Adaptation with Masked Pre-Training and Collaborative Self-Training (CVPR 2024 · similarity 0.87)
- Learning Causal Domain-Invariant Temporal Dynamics for Few-Shot Action Recognition (ICML 2024 · similarity 0.86)
- ActionAtlas: A VideoQA Benchmark for Domain-specialized Action Recognition (NeurIPS 2024 · similarity 0.86)
- Align Before Adapt: Leveraging Entity-to-Region Alignments for Generalizable Video Action Recognition (CVPR 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.
