Automatic Prompt Generation Using Class Activation Maps for Foundational Models: A Polyp Segmentation Case Study

We introduce a weakly supervised segmentation approach that leverages class activation maps and the Segment Anything Model to generate high-quality masks using only classification data. A pre-trained classifier produces class activation maps that, once thresholded, yield bounding boxes encapsulating...

Full description

Saved in:
Bibliographic Details
Main Authors: Hanna Borgli, Håkon Kvale Stensland, Pål Halvorsen
Format: Article
Language:English
Published: MDPI AG 2025-02-01
Series:Machine Learning and Knowledge Extraction
Subjects:
Online Access:https://www.mdpi.com/2504-4990/7/1/22
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850204976643571712
author Hanna Borgli
Håkon Kvale Stensland
Pål Halvorsen
author_facet Hanna Borgli
Håkon Kvale Stensland
Pål Halvorsen
author_sort Hanna Borgli
collection DOAJ
description We introduce a weakly supervised segmentation approach that leverages class activation maps and the Segment Anything Model to generate high-quality masks using only classification data. A pre-trained classifier produces class activation maps that, once thresholded, yield bounding boxes encapsulating the regions of interest. These boxes prompt the SAM to generate detailed segmentation masks, which are then refined by selecting the best overlap with automatically generated masks from the foundational model using the intersection over union metric. In a polyp segmentation case study, our approach outperforms existing zero-shot and weakly supervised methods, achieving a mean intersection over union of 0.63. This method offers an efficient and general solution for image segmentation tasks where segmentation data are scarce.
format Article
id doaj-art-c457f5057fc34ad79512430879bf1b85
institution OA Journals
issn 2504-4990
language English
publishDate 2025-02-01
publisher MDPI AG
record_format Article
series Machine Learning and Knowledge Extraction
spelling doaj-art-c457f5057fc34ad79512430879bf1b852025-08-20T02:11:11ZengMDPI AGMachine Learning and Knowledge Extraction2504-49902025-02-01712210.3390/make7010022Automatic Prompt Generation Using Class Activation Maps for Foundational Models: A Polyp Segmentation Case StudyHanna Borgli0Håkon Kvale Stensland1Pål Halvorsen2Department of High-Performance Computing, Simula Research Laboratory, 0164 Oslo, NorwayDepartment of High-Performance Computing, Simula Research Laboratory, 0164 Oslo, NorwayDepartment of Holistic Systems, SimulaMet, 0170 Oslo, NorwayWe introduce a weakly supervised segmentation approach that leverages class activation maps and the Segment Anything Model to generate high-quality masks using only classification data. A pre-trained classifier produces class activation maps that, once thresholded, yield bounding boxes encapsulating the regions of interest. These boxes prompt the SAM to generate detailed segmentation masks, which are then refined by selecting the best overlap with automatically generated masks from the foundational model using the intersection over union metric. In a polyp segmentation case study, our approach outperforms existing zero-shot and weakly supervised methods, achieving a mean intersection over union of 0.63. This method offers an efficient and general solution for image segmentation tasks where segmentation data are scarce.https://www.mdpi.com/2504-4990/7/1/22image segmentationclass activation mapsegment anythingannotation toolszero-shot learningweakly supervised semantic segmentation
spellingShingle Hanna Borgli
Håkon Kvale Stensland
Pål Halvorsen
Automatic Prompt Generation Using Class Activation Maps for Foundational Models: A Polyp Segmentation Case Study
Machine Learning and Knowledge Extraction
image segmentation
class activation map
segment anything
annotation tools
zero-shot learning
weakly supervised semantic segmentation
title Automatic Prompt Generation Using Class Activation Maps for Foundational Models: A Polyp Segmentation Case Study
title_full Automatic Prompt Generation Using Class Activation Maps for Foundational Models: A Polyp Segmentation Case Study
title_fullStr Automatic Prompt Generation Using Class Activation Maps for Foundational Models: A Polyp Segmentation Case Study
title_full_unstemmed Automatic Prompt Generation Using Class Activation Maps for Foundational Models: A Polyp Segmentation Case Study
title_short Automatic Prompt Generation Using Class Activation Maps for Foundational Models: A Polyp Segmentation Case Study
title_sort automatic prompt generation using class activation maps for foundational models a polyp segmentation case study
topic image segmentation
class activation map
segment anything
annotation tools
zero-shot learning
weakly supervised semantic segmentation
url https://www.mdpi.com/2504-4990/7/1/22
work_keys_str_mv AT hannaborgli automaticpromptgenerationusingclassactivationmapsforfoundationalmodelsapolypsegmentationcasestudy
AT hakonkvalestensland automaticpromptgenerationusingclassactivationmapsforfoundationalmodelsapolypsegmentationcasestudy
AT palhalvorsen automaticpromptgenerationusingclassactivationmapsforfoundationalmodelsapolypsegmentationcasestudy