Research map / Detection and segmentation / Open-set and few-shot detection
SPIN: Hierarchical Segmentation with Subpart Granularity in Natural Images
Josh David Myers-Dean, Jarek T Reynolds, Brian Price, Yifei Fan, Danna Gurari
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 |
|---|---|---|---|
| fine-tune GLaMM on SPIN training data | GLaMM 7B · SPIN subparts (General prompt) | mIoU_S: 11.00 -> 24.56 | Improved |
| fine-tune GLaMM on SPIN training data | GLaMM 7B · SPIN parts (General prompt) | mIoU_P: 40.00 -> 60.76 | Improved |
| fine-tune GLaMM on SPIN training data | GLaMM 7B · SPIN objects (General prompt) | mIoU_O: 86.31 -> 91.08 | Improved |
| use hierarchical-specific HIPIE instead of general foundation models | hierarchical segmentation · SPIN subpart localization | mIoU_S: 0.92 (HIPIE ViT-H) vs 11.55 (LISA 13B) | Hurt |
| prompt with general rather than specific object categories (ablation) | CoGVLM, LISA · SPIN objects | mIoU_O: -2.82pp CoGVLM; -2.33pp LISA-13B; -2.87pp LISA-7B | Hurt |
| general versus specific prompts for parts and subparts (ablation) | foundation models · SPIN parts and subparts | mIoU: gap within 1pp | No clear effect |
| give SAM ground-truth boxes instead of language-only prompting (versus GLaMM) | SAM · SPIN subparts | mIoU_S: +38.61pp subparts, +20.23pp parts, +3.75pp objects | Improved |
| pair SAM with CoGVLM predicted boxes versus GLaMM | SAM + CoGVLM · SPIN subparts, parts and objects | mIoU: +9.24pp subparts, +10.87pp parts, -11.22pp objects | Mixed |
| move from object to part to subpart granularity | all models · SPIN zero-shot localization | mIoU: 86 -> 40 -> 14 (best object, part, subpart) | Hurt |
| scale ViP-Llava from 7B to 13B for recognition | ViP-Llava · SPIN recognition | mACC_S (%): 53.57 -> 94.78 | Improved |
Closest papers by meaning
- Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure (NeurIPS 2024 · similarity 0.89)
- Flattening the Parent Bias: Hierarchical Semantic Segmentation in the Poincare Ball (CVPR 2024 · similarity 0.88)
- Learning Hierarchical Image Segmentation For Recognition and By Recognition (ICLR 2024 · similarity 0.88)
- Hierarchical Prototype Learning for Semantic Segmentation (ICLR 2026 · similarity 0.86)
- Unsupervised Hierarchical Semantic Segmentation With Multiview Cosegmentation and Clustering Transformers (CVPR 2022 · similarity 0.86)
- Semantic Hierarchy-Aware Segmentation (TPAMI 2023 · 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.
