Papers › Mitigating Background Shift in Class-Incremental Semantic Segmentation

Mitigating Background Shift in Class-Incremental Semantic Segmentation

16 Jul 2024arXiv:2407.11859archive 2025-07-28

Gilhan Park, WonJun Moon, SuBeen Lee, Tae-Young Kim, Jae-Pil Heo

Class-Incremental Semantic Segmentation(CISS) aims to learn new classes without forgetting the old ones, using only the labels of the new classes. To achieve this, two popular strategies are employed: 1) pseudo-labeling and knowledge distillation to preserve prior knowledge; and 2) background weight transfer, which leverages the broad coverage of background in learning new classes by transferring background weight to the new class classifier. However, the first strategy heavily relies on the old model in detecting old classes while undetected pixels are regarded as the background, thereby leading to the background shift towards the old classes(i.e., misclassification of old class as background). Additionally, in the case of the second approach, initializing the new class classifier with background knowledge triggers a similar background shift issue, but towards the new classes. To address these issues, we propose a background-class separation framework for CISS. To begin with, selective pseudo-labeling and adaptive feature distillation are to distill only trustworthy past knowledge. On the other hand, we encourage the separation between the background and new classes with a novel orthogonal objective along with label-guided output distillation. Our state-of-the-art results validate the effectiveness of these proposed methods.

PaperPDFCode

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

Code

roadonep/eccv2024_mbs officialmentioned in papermentioned 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

Class Incremental LearningClass-Incremental Semantic SegmentationContinual LearningContinual Semantic SegmentationDisjoint 15-1Disjoint 15-5Disjoint 19-1Knowledge DistillationOverlapped 10-1Overlapped 100-10Overlapped 100-5Overlapped 100-50Overlapped 15-1Overlapped 15-5Overlapped 5-3Overlapped 50-50SegmentationSemantic Segmentation

1 archive task tag without a task page not shown.

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Disjoint 15-1 PASCAL VOC 2012 MBS mIoU 78.1 #1 of 9 Archive leaderboard report
Disjoint 15-5 PASCAL VOC 2012 MBS Mean IoU 79.0 #1 of 9 Archive leaderboard report
Disjoint 19-1 PASCAL VOC 2012 MBS mIoU 82.8 #1 of 1 Archive leaderboard report
Overlapped 10-1 PASCAL VOC 2012 MBS mIoU 77.19 #1 of 13 Archive leaderboard report
Overlapped 100-10 ADE20K MBS Mean IoU (test) 44.5 #1 of 6 Archive leaderboard report
Overlapped 100-5 ADE20K MBS mIoU 42.8 #1 of 8 Archive leaderboard report
Overlapped 100-50 ADE20K MBS mIoU 45.7 #1 of 7 Archive leaderboard report
Overlapped 15-1 PASCAL VOC 2012 MBS mIoU 80.6 #1 of 13 Archive leaderboard report
Overlapped 15-5 PASCAL VOC 2012 MBS Mean IoU (val) 82.6 #1 of 13 Archive leaderboard report
Overlapped 5-3 PASCAL VOC 2012 MBS Mean IoU (test) 78.1 #1 of 4 Archive leaderboard report
Overlapped 50-50 ADE20K MBS mIoU 45.4 #1 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.

Methods

Knowledge Distillation

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