Research map / Detection and segmentation / Weakly supervised detection and segmentation
Block Annotation: Better Image Annotation With Sub-Image Decomposition
Hubert Lin, Paul Upchurch, Kavita Bala
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 |
|---|---|---|---|
| annotate 50% of blocks per image instead of full images | full annotation · Cityscapes val (all images) | mIoU: 77.7 -> 77.7 | No clear effect |
| annotate 50% of blocks per image instead of full images | full annotation · ADE20K val (all images) | mIoU: 37.4 -> 37.2 | No clear effect |
| annotate 12% of blocks per image instead of full images | full annotation · Cityscapes val (all images) | mIoU: 77.7 -> 74.6 | Hurt |
| annotate 12% of blocks per image instead of full images | full annotation · ADE20K val (all images) | mIoU: 37.4 -> 36.1 | Hurt |
| use block annotation instead of full-image annotation at equal pixel budget | full annotation · Cityscapes and ADE20K | error reduction: 13% (10%) error reduction at 12% budget | Improved |
| use pseudo-checkerboard block selection instead of checkerboard (ablation) | checkerboard blocks · Cityscapes and ADE20K | mIoU | Improved |
| use block annotation instead of scribbles at equal time | ScribbleSup · Pascal VOC 2012 val | mIoU: 63.1 -> 67.2 | Improved |
| use block annotation instead of coarse annotation at equal time | Coarse annotation · Cityscapes val | mIoU: 68.8 -> 72.1 | Improved |
| use block annotation instead of full-image annotation for crowdsourcing | Full-image annotation · SUNCG annotation error | error rate: 0.286 -> 0.253 | Improved |
| use block annotation instead of full-image annotation for crowdsourcing | Full-image annotation · SUNCG annotation error (small regions) | error rate: 0.677 -> 0.636 | Improved |
| block inpainting instead of automatic segmentation | Automatic segmentation · Cityscapes val (Block-50%) | mIoU: 78 -> 92 | Improved |
| block inpainting instead of automatic segmentation | Automatic segmentation · Cityscapes val (Block-12%) | mIoU: 75 -> 82 | Improved |
Closest papers by meaning
- Full or Weak Annotations? An Adaptive Strategy for Budget-Constrained Annotation Campaigns (CVPR 2023 · similarity 0.88)
- Joint Inference in Weakly-Annotated Image Datasets via Dense Correspondence (IJCV 2016 · similarity 0.88)
- What’s the Point: Semantic Segmentation with Point Supervision (ECCV 2016 · similarity 0.87)
- Devil Is in the Edges: Learning Semantic Boundaries From Noisy Annotations (CVPR 2019 · similarity 0.87)
- Scaling Up Instance Annotation via Label Propagation (ICCV 2021 · similarity 0.87)
- Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure (NeurIPS 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.
