Patch-sampled contrastive learning for dense prediction pretraining in metallographic images
Mingchun Li, Yang Liu, Dali Chen, Jining Bao, Yan Huo · 2025
Read Patch-sampled contrastive learning for dense prediction pretraining in metallographic images by Mingchun Li, Yang Liu, Dali Chen, Jining Bao, Yan Huo (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-12830817.txt.
Description
Mingchun Li and co-authors pretrain image models for metal microstructure analysis without costly labels, in this 2025 Europe PMC paper. Microstructure shapes in alloy images decide mechanical properties, but annotating them is expensive. Their method, patch-sampled contrastive learning, learns from the images themselves first.
Contrastive training runs at two scales: whole images for global features and small patches for local detail, with a multiscale strategy for later segmentation. Because one patch can hold several microstructures, sampling is guided by feature similarity to keep categories separable. Fine-tuned on a single annotated image, the model beat rival self-supervised methods at segmenting microstructures.
Catalog record
| CREATOR | Mingchun Li, Yang Liu, Dali Chen, Jining Bao, Yan Huo |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 82 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |