Research map / Video / Action recognition
End-to-End Joint Semantic Segmentation of Actors and Actions in Video
Jingwei Ji, Shyamal Buch, Alvaro Soto, Juan Carlos Niebles
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 decoupled end-to-end model instead of single-frame Mask R-CNN on cross-product labels (ablation) | actor-action segmentation · A2D | joint actor-action mIoU: 19.1 -> 36.9 | Improved |
| add mask head end-to-end instead of two-stage boxes plus SharpMask (ablation) | actor-action segmentation · A2D | joint actor-action mIoU: 32.4 -> 36.9 | Improved |
| add temporal 3D conv layers (ablation) | actor-action segmentation · A2D | action mIoU: 45.5 -> 46.3 | Improved |
| add optical flow stream (ablation) | actor-action segmentation · A2D | action mIoU: 36.8 -> 46.3 | Improved |
| add flow stream to actor segmentation (ablation) | actor-action segmentation · A2D | actor mIoU: 66.5 -> 66.4 | No clear effect |
| train actor and action branches jointly instead of separately (ablation) | actor-action segmentation · A2D | joint actor-action mIoU: 33.6 -> 36.9 | Improved |
| use end-to-end joint model instead of TSMT + SM | actor-action segmentation · A2D | joint actor-action mIoU: 29.7 -> 36.9 | Improved |
| use end-to-end model instead of TSMT for detection | actor-action detection · A2D spatial detection | joint actor-action mAP: 48.9 -> 55.5 | Improved |
| use appearance-only RoIs instead of appearance plus motion RoIs | RPN · A2D | RoI quality | Improved |
| use mid-level fusion instead of late fusion of appearance and motion | two-stream action recognition · A2D | segmentation | Improved |
Closest papers by meaning
- Actor and Action Video Segmentation From a Sentence (CVPR 2018 · similarity 0.90)
- Actor-Action Semantic Segmentation With Grouping Process Models (CVPR 2016 · similarity 0.90)
- Context Modulated Dynamic Networks for Actor and Action Video Segmentation with Language Queries (AAAI 2020 · similarity 0.89)
- Weakly Supervised Actor-Action Segmentation via Robust Multi-Task Ranking (CVPR 2017 · similarity 0.89)
- Unsupervised Action Segmentation via Fast Learning of Semantically Consistent Actoms (AAAI 2024 · similarity 0.88)
- FACT: Frame-Action Cross-Attention Temporal Modeling for Efficient Action Segmentation (CVPR 2024 · similarity 0.87)
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.
