Physics Maths Engineering
Hui Fang,
Kunhao Yuan,
Gerald Schaefer,
Yu-Kun Lai,
Yifan Wang,
Xiyao Liu
Peer Reviewed
Weakly supervised semantic segmentation (WSSS) has gained significant popularity as it relies only on weak labels such as image level annotations rather than the pixel level annotations required by supervised semantic segmentation (SSS) methods. Despite drastically reduced annotation costs, typical feature representations learned from WSSS are only representative of some salient parts of objects and less reliable compared to SSS due to the weak guidance during training. In this paper, we propose a novel Multi-Strategy Contrastive Learning (MuSCLe) framework to obtain enhanced feature representations and improve WSSS performance by exploiting similarity and dissimilarity of contrastive sample pairs at image, region, pixel and object boundary levels. Extensive experiments demonstrate the effectiveness of our method and show that MuSCLe outperforms current state-of-the-art methods on the widely used PASCAL VOC 2012 dataset.
WSSS is a technique for segmenting images into meaningful regions (e.g., objects) using only weak labels, such as image-level annotations, instead of expensive pixel-level annotations. It reduces annotation costs while still achieving competitive results.
The main challenge is that WSSS often learns feature representations that focus only on the most salient parts of objects, missing finer details. This makes it less reliable compared to fully supervised methods that use pixel-level annotations.
MuSCLe is a novel framework that improves WSSS by leveraging contrastive learning at multiple levels: image, region, pixel, and object boundaries. It enhances feature representations by exploiting similarities and differences between contrastive sample pairs.
MuSCLe improves WSSS by:
MuSCLe applies contrastive learning at four levels:
MuSCLe was tested on the widely used PASCAL VOC 2012 dataset. Experiments showed that it outperforms current state-of-the-art WSSS methods, demonstrating its effectiveness in improving segmentation accuracy.
MuSCLe:
MuSCLe can be used in:
While fully supervised methods require pixel-level annotations, MuSCLe achieves competitive results using only image-level labels, significantly reducing annotation costs. It bridges the gap between weakly supervised and fully supervised approaches.
Future research could explore:
Show by month | Manuscript | Video Summary |
---|---|---|
2025 April | 2 | 2 |
2025 March | 75 | 75 |
2025 February | 45 | 45 |
2025 January | 44 | 44 |
2024 December | 47 | 47 |
2024 November | 43 | 43 |
2024 October | 35 | 35 |
2024 September | 56 | 56 |
2024 August | 37 | 37 |
2024 July | 33 | 33 |
2024 June | 27 | 27 |
2024 May | 35 | 35 |
2024 April | 22 | 22 |
2024 March | 6 | 6 |
Total | 507 | 507 |
Show by month | Manuscript | Video Summary |
---|---|---|
2025 April | 2 | 2 |
2025 March | 75 | 75 |
2025 February | 45 | 45 |
2025 January | 44 | 44 |
2024 December | 47 | 47 |
2024 November | 43 | 43 |
2024 October | 35 | 35 |
2024 September | 56 | 56 |
2024 August | 37 | 37 |
2024 July | 33 | 33 |
2024 June | 27 | 27 |
2024 May | 35 | 35 |
2024 April | 22 | 22 |
2024 March | 6 | 6 |
Total | 507 | 507 |