Research map / Detection and segmentation / Weakly supervised detection and segmentation
Weakly-Supervised Image Annotation and Segmentation with Objects and Attributes
Zhiyuan Shi, Yongxin Yang, Timothy M. Hospedales, Tao Xiang
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 8 material attribute annotations to weakly supervised segmentation (ablation) | WS-MRF-SIBP · aPascal segmentation | avg IOU (%): 23.6 -> 27.3 | Improved |
| use 64 attributes instead of 8 (ablation) | WS-MRF-SIBP · aPascal segmentation | avg IOU (%): 27.3 -> 28.9 | Improved |
| use 74 attributes (add 10 color attributes) instead of 64 (ablation) | WS-MRF-SIBP · aPascal segmentation | avg IOU (%): 28.9 -> 29.4 | Improved |
| replace WSDC with WS-MRF-SIBP without attributes | weakly supervised segmentation · aPascal segmentation | avg IOU (%): 18.2 -> 23.6 | Improved |
| update the model with test images (transductive mode) (ablation) | WS-MRF-SIBP · LabelMe Outdoor | per-pixel accuracy (%): 46.2 -> 52.5 | Improved |
| update the model with test images (transductive mode) (ablation) | WS-MRF-SIBP · LabelMe Outdoor | per-class accuracy (%): 23.8 -> 31.2 | Improved |
| use image-level labels predicted by an external CNN+SVM classifier instead of the model's own predictions (ablation) | WS-MRF-SIBP transductive · LabelMe Outdoor | per-pixel accuracy (%): 52.5 -> 48.1 | Hurt |
| replace w-LDA with WS-MRF-SIBP (IBP, non-competing factors) | free annotation · aPascal free annotation | AP@2: 30.7 -> 40.1 | Improved |
| replace weak labels with strong bounding-box supervision | annotation · ImageNet Attribute free annotation | AP@2: 60.7 -> 65.9 | Improved |
| use weak supervision instead of strong supervision for attribute annotation given locations | WS-MRF-SIBP · aPascal annotation given locations | mAP: 42.1 -> 43.6 | No clear effect |
| use WS-MRF-SIBP instead of w-LDA | training cost · aPascal training set | seconds per iteration: 65 -> 80 | Hurt |
Closest papers by meaning
- Joint Inference in Weakly-Annotated Image Datasets via Dense Correspondence (IJCV 2016 · similarity 0.90)
- Universal Weakly Supervised Segmentation by Pixel-to-Segment Contrastive Learning (ICLR 2021 · similarity 0.88)
- Weakly- and Semi-Supervised Panoptic Segmentation (ECCV 2018 · similarity 0.88)
- Looking Beyond Single Images for Weakly Supervised Semantic Segmentation Learning (TPAMI 2022 · similarity 0.88)
- Combining Bottom-Up, Top-Down, and Smoothness Cues for Weakly Supervised Image Segmentation (CVPR 2017 · similarity 0.88)
- Relational Matching for Weakly Semi-Supervised Oriented Object Detection (CVPR 2024 · 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.
