Papers › DiscoBox: Weakly Supervised Instance Segmentation and Semantic Correspondence from Box...

DiscoBox: Weakly Supervised Instance Segmentation and Semantic Correspondence from Box Supervision

13 May 2021ICCV 2021 10arXiv:2105.06464archive 2025-07-28

Shiyi Lan, Zhiding Yu, Christopher Choy, Subhashree Radhakrishnan, Guilin Liu, Yuke Zhu, Larry S. Davis, Anima Anandkumar

We introduce DiscoBox, a novel framework that jointly learns instance segmentation and semantic correspondence using bounding box supervision. Specifically, we propose a self-ensembling framework where instance segmentation and semantic correspondence are jointly guided by a structured teacher in addition to the bounding box supervision. The teacher is a structured energy model incorporating a pairwise potential and a cross-image potential to model the pairwise pixel relationships both within and across the boxes. Minimizing the teacher energy simultaneously yields refined object masks and dense correspondences between intra-class objects, which are taken as pseudo-labels to supervise the task network and provide positive/negative correspondence pairs for dense constrastive learning. We show a symbiotic relationship where the two tasks mutually benefit from each other. Our best model achieves 37.9% AP on COCO instance segmentation, surpassing prior weakly supervised methods and is competitive to supervised methods. We also obtain state of the art weakly supervised results on PASCAL VOC12 and PF-PASCAL with real-time inference.

PaperPDFConference PDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

NVlabs/DiscoBox mentioned on GitHubpytorch report
voidrank/DiscoBox mentioned on GitHubpytorch report
voidrank/SCOT mentioned on GitHubpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Box-supervised Instance SegmentationInstance SegmentationSegmentationSemantic correspondenceWeakly-supervised instance segmentation

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Box-supervised Instance Segmentation COCO test-dev DiscoBox mask AP 37.9 #3 of 7 Archive leaderboard report
Box-supervised Instance Segmentation PASCAL VOC 2012 val DiscoBox AP_25 75.2 #5 of 5 Archive leaderboard report
Box-supervised Instance Segmentation PASCAL VOC 2012 val DiscoBox AP_50 63.6 #5 of 5 Archive leaderboard report
Box-supervised Instance Segmentation PASCAL VOC 2012 val DiscoBox AP_70 45.5 #5 of 5 Archive leaderboard report
Box-supervised Instance Segmentation PASCAL VOC 2012 val DiscoBox AP_75 37.5 #5 of 5 Archive leaderboard report
Weakly-supervised instance segmentation COCO 2017 val DiscoBox (ResNet-50) AP 31.4 #1 of 2 Archive leaderboard report
Weakly-supervised instance segmentation COCO 2017 val DiscoBox (ResNet-50) AP@50 52.6 #1 of 2 Archive leaderboard report
Weakly-supervised instance segmentation COCO 2017 val DiscoBox (ResNet-50) AP@75 32.2 #1 of 2 Archive leaderboard report
Weakly-supervised instance segmentation COCO 2017 val DiscoBox (ResNet-50) AP@L 50.1 #1 of 2 Archive leaderboard report
Weakly-supervised instance segmentation COCO 2017 val DiscoBox (ResNet-50) AP@M 33.8 #1 of 2 Archive leaderboard report
Weakly-supervised instance segmentation COCO 2017 val DiscoBox (ResNet-50) AP@S 11.5 #1 of 2 Archive leaderboard report
Weakly-supervised instance segmentation COCO test-dev DiscoBox (ResNeXt-101-DCN-FPN) AP 37.9 #1 of 7 Archive leaderboard report
Weakly-supervised instance segmentation COCO test-dev DiscoBox (ResNeXt-101-DCN-FPN) AP@50 61.4 #1 of 7 Archive leaderboard report
Weakly-supervised instance segmentation COCO test-dev DiscoBox (ResNeXt-101-DCN-FPN) AP@75 40.0 #1 of 7 Archive leaderboard report
Weakly-supervised instance segmentation COCO test-dev DiscoBox (ResNeXt-101-DCN-FPN) AP@L 53.9 #1 of 7 Archive leaderboard report
Weakly-supervised instance segmentation COCO test-dev DiscoBox (ResNeXt-101-DCN-FPN) AP@M 41.1 #1 of 7 Archive leaderboard report
Weakly-supervised instance segmentation COCO test-dev DiscoBox (ResNeXt-101-DCN-FPN) AP@S 18.0 #1 of 7 Archive leaderboard report
Weakly-supervised instance segmentation COCO test-dev DiscoBox (ResNet-101-DCN-FPN) AP 35.8 #2 of 7 Archive leaderboard report
Weakly-supervised instance segmentation COCO test-dev DiscoBox (ResNet-101-DCN-FPN) AP@50 59.8 #2 of 7 Archive leaderboard report
Weakly-supervised instance segmentation COCO test-dev DiscoBox (ResNet-101-DCN-FPN) AP@75 36.4 #2 of 7 Archive leaderboard report
Weakly-supervised instance segmentation COCO test-dev DiscoBox (ResNet-101-DCN-FPN) AP@L 52.1 #2 of 7 Archive leaderboard report
Weakly-supervised instance segmentation COCO test-dev DiscoBox (ResNet-101-DCN-FPN) AP@M 38.7 #2 of 7 Archive leaderboard report
Weakly-supervised instance segmentation COCO test-dev DiscoBox (ResNet-101-DCN-FPN) AP@S 16.9 #2 of 7 Archive leaderboard report
Weakly-supervised instance segmentation COCO test-dev DiscoBox (ResNet-50-FPN) AP 32.0 #7 of 7 Archive leaderboard report
Weakly-supervised instance segmentation COCO test-dev DiscoBox (ResNet-50-FPN) AP@50 53.6 #7 of 7 Archive leaderboard report
Weakly-supervised instance segmentation COCO test-dev DiscoBox (ResNet-50-FPN) AP@75 32.6 #7 of 7 Archive leaderboard report
Weakly-supervised instance segmentation COCO test-dev DiscoBox (ResNet-50-FPN) AP@L 48.4 #7 of 7 Archive leaderboard report
Weakly-supervised instance segmentation COCO test-dev DiscoBox (ResNet-50-FPN) AP@M 33.7 #7 of 7 Archive leaderboard report
Weakly-supervised instance segmentation COCO test-dev DiscoBox (ResNet-50-FPN) AP@S 11.7 #7 of 7 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections