Research map / Detection and segmentation / Weakly supervised detection and segmentation
Learning to Segment Human by Watching YouTube
Xiaodan Liang, Yunchao Wei, Liang Lin, Yunpeng Chen, Xiaohui Shen, Jianchao Yang, Shuicheng Yan
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 video-context guided mask inference over image-based segmentation (ablation) | iterative framework · VOC 2012 val (person) | IoU (%): 72.9 -> 81.8 | Improved |
| add pairwise potentials (ablation) | video-context inference · VOC 2012 val (person) | IoU (%): 80.5 -> 81.8 | Improved |
| add higher-order potentials (ablation) | video-context inference · VOC 2012 val (person) | IoU (%): 80.9 -> 81.8 | Improved |
| iterate training and mask refinement for 10 iterations (ablation) | Ours · VOC 2012 val (person) | IoU (%): 28.1 -> 81.8 | Improved |
| add sample-weighted loss (ablation) | Ours · VOC 2012 val (person) | IoU (%): 76.8 -> 81.8 | Improved |
| add more negative frames from raw videos (ablation) | Ours · VOC 2012 val (person) | IoU (%): 74.9 -> 81.8 | Improved |
| add sample-weighted loss and negative images together (ablation) | Ours · VOC 2012 val (person) | IoU (%): 70.3 -> 81.8 | Improved |
| add 10k bounding box annotations in last iteration | Ours · VOC 2012 val (person) | IoU (%): +0.4% | No clear effect |
| add 10k mask annotations in last iteration | Ours · VOC 2012 val (person) | IoU (%): 81.5 -> 83.0 | Improved |
| use EM procedure to drop frames with few foreground pixels (ablation) | image-level person-label training · VOC 2012 val (person) | IoU (%): 34.2 -> 52.8 | Improved |
| use video-context inference instead of image co-localization boxes | Ours · VOC 2012 val (person) | IoU (%): 40.6 -> 81.5 | Improved |
| train on 2-class person task instead of multi-class (ablation) | DeepLab-CRF · VOC 2012 test (person) | IoU (%): 77.6 -> 76.7 | Hurt |
Closest papers by meaning
- Weakly Supervised Instance Segmentation for Videos With Temporal Mask Consistency (CVPR 2021 · similarity 0.88)
- Fine-Grained Human-Centric Tracklet Segmentation with Single Frame Supervision (TPAMI 2019 · similarity 0.87)
- Leveraging Motion Priors in Videos for Improving Human Segmentation (ECCV 2018 · similarity 0.87)
- YouTube-BoundingBoxes: A Large High-Precision Human-Annotated Data Set for Object Detection in Video (CVPR 2017 · similarity 0.87)
- Unified Mask Embedding and Correspondence Learning for Self-Supervised Video Segmentation (CVPR 2023 · similarity 0.86)
- Weakly-Supervised Semantic Segmentation Using Motion Cues (ECCV 2016 · similarity 0.86)
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.
