I know what contrastive learning is, and I know what other traditional segmentation losses are. What I understand is that the goal of contrastive loss is basically to pull similar things together and push dissimilar things apart. But I want to know how this can guide a segmentation pipeline (e.g. semantic segmentation)? My question is pretty basic. Blogs/video links are more welcome than research paper links.
This would be a great question to post on imaging QA, a free medical imaging Question and Answer community: www.imagingQA.com
@george
Thank you for your response. However, if you notice there are plenty of contrastive learning based semantic segmentation research going on. InfoNCE, SIMCLR, JCL being some of the well-known paradigms. Some of the methods utilise contrastive learning for self-supervised network pretraining, which is further used for segmentation. I don't understand why this works. I completely agree with your answer and the effectiveness of CL in classification. But thinking of some idea in this direction. That's why I asked.
George C. Giakos
Thank you for your response. However, if you notice there are plenty of contrastive learning based semantic segmentation research going on. InfoNCE, SIMCLR, JCL being some of the well-known paradigms. Some of the methods utilise contrastive learning for self-supervised network pretraining, which is further used for segmentation. I don't understand why this works. I completely agree with your answer and the effectiveness of CL in classification. But thinking of some idea in this direction. That's why I asked.
- Badges
- Science topic
- Similar topics
- Methodology
- Research Methodology
- Research Papers