Research map / Vision-language and multimodal / CLIP and zero-shot
Text Promptable Surgical Instrument Segmentation with Vision-Language Models
Zijian Zhou, Oluwatosin Alabi, Meng Wei, Tom Vercauteren, Miaojing Shi
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 |
|---|---|---|---|
| remove multi-scale feature augmentation (MSFA) (ablation) | text promptable segmentation model · EndoVis2017 (448) | Ch_IoU: 75.31 -> 77.79 | Improved |
| replace [CLS] text feature with averaged word tokens (ablation) | text promptable segmentation model · EndoVis2017 (448) | Ch_IoU: 77.33 -> 77.79 | Improved |
| use only convolutional prompting instead of attention then convolution (ablation) | mask decoder · EndoVis2017 (448) | Ch_IoU: 42.38 -> 77.79 | Improved |
| use only attention-based prompting (ablation) | mask decoder · EndoVis2017 (448) | Ch_IoU: 76.46 -> 77.79 | Improved |
| use only class prompts instead of mixture of prompts (ablation) | MoP · EndoVis2017 (448) | Ch_IoU: 75.67 -> 77.79 | Improved |
| use only GPT-4 prompt instead of full mixture (ablation) | MoP · EndoVis2017 (448) | Ch_IoU: 77.19 -> 77.79 | Improved |
| replace class prompt with Bard-generated prompt (ablation) | MoP · EndoVis2017 (448) | Ch_IoU: 77.79 -> 77.76 | No clear effect |
| replace pixel-wise gating weight maps with image-wise scalar weights (ablation) | MoP gating network · EndoVis2017 (448) | Ch_IoU: 76.80 -> 77.79 | Improved |
| remove hard instrument area reinforcement (HIAR) (ablation) | text promptable segmentation model · EndoVis2017 (448) | Ch_IoU: 75.22 -> 77.79 | Improved |
| remove hard area mining (HAM) from HIAR (ablation) | HIAR · EndoVis2017 (448) | Ch_IoU: 75.98 -> 77.79 | Improved |
| increase input size from 448 to 896 | text promptable segmentation model · EndoVis2017/2018 | mc_IoU: 54.78 -> 56.22 | Mixed |
| cross-dataset testing (train on EndoVis2018, test on EndoVis2017) | text promptable segmentation model · EndoVis2017 | Ch_IoU: 79.90 -> 72.18 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CRIS — CRIS: CLIP-Driven Referring Image Segmentation (CVPR 2022) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SurgicalSAM: Efficient Class Promptable Surgical Instrument Segmentation (AAAI 2024 · similarity 0.92)
- CAT: Coordinating Anatomical-Textual Prompts for Multi-Organ and Tumor Segmentation (NeurIPS 2024 · similarity 0.85)
- Recognize Any Surgical Object: Unleashing the Power of Weakly-Supervised Data (ICLR 2025 · similarity 0.83)
- Segment Anyword: Mask Prompt Inversion for Open-Set Grounded Segmentation (ICML 2025 · similarity 0.82)
- BiomedCoOp: Learning to Prompt for Biomedical Vision-Language Models (CVPR 2025 · similarity 0.82)
- Procedure-Aware Surgical Video-language Pretraining with Hierarchical Knowledge Augmentation (NeurIPS 2024 · similarity 0.82)
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.
