Research map / Detection and segmentation / Open-set and few-shot detection
Deep Extreme Cut: From Extreme Points to Object Segmentation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 Deeplab-v2 instead of region-based ResNet-101-C4 (ablation) | DEXTR architecture · PASCAL VOC 2012 val | IoU: +3.9% | Helped |
| use extreme points instead of bounding boxes (ablation) | Deeplab-v2 input guidance · PASCAL VOC 2012 val | IoU: +3.1% | Helped |
| use balanced BCE instead of standard cross-entropy (ablation) | DEXTR loss · PASCAL VOC 2012 val | IoU: +3.3% | Helped |
| crop around the object with 50-pixel context instead of processing the full image (ablation) | DEXTR input · PASCAL VOC 2012 val | IoU: +7.9% | Helped |
| use PSP instead of ASPP (ablation) | DEXTR network head · PASCAL VOC 2012 val | IoU: +2.3% | Helped |
| replace fixed Gaussian point maps with a distance transform (ablation) | DEXTR point representation · PASCAL VOC 2012 val | IoU: -1.3% | Hurt |
| add PSP to cropped Deeplab-v2 (ablation) | DEXTR · PASCAL VOC 2012 val | IoU (%): 85.1 -> 87.4; +2.3% | Helped |
| add extreme points to cropped Deeplab-v2 with PSP (ablation) | DEXTR · PASCAL VOC 2012 val | IoU (%): 87.4 -> 90.5; +3.1% | Helped |
| add SBD data to the cropped Deeplab-v2 + PSP + extreme-points method (ablation) | DEXTR · PASCAL VOC 2012 val | IoU (%): 90.5 -> 91.5; +1.0% | Helped |
| use manual rather than simulated extreme points (ablation) | DEXTR · PASCAL_EXT and DAVIS 2016 | IoU: PASCAL_EXT: 85.1 -> 80.1; DAVIS 2016: 79.5 -> 80.9 | Mixed |
Closest papers by meaning
- FastMask: Segment Multi-Scale Object Candidates in One Shot (CVPR 2017 · similarity 0.88)
- Hard ImageNet: Segmentations for Objects with Strong Spurious Cues (NeurIPS 2022 · similarity 0.88)
- Putting the Object Back into Video Object Segmentation (CVPR 2024 · similarity 0.87)
- Extreme Point Supervised Instance Segmentation (CVPR 2024 · similarity 0.87)
- unMORE: Unsupervised Multi-Object Segmentation via Center-Boundary Reasoning (ICML 2025 · similarity 0.86)
- Learning Hierarchical Image Segmentation For Recognition and By Recognition (ICLR 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.