A small project about corrosion detection (course project). The pipeline is as follows:
For a given image, use resnet
for image classification. If it's classified as rust, use unet
to get the corrosion mask.
All the images from classification dataset are collected from internet (some of them are from Oxford pet datasets). The small dataset for corrosion segmentation is annotated by myself (Annotation tool). Because of the lack of large dataset, the performance might be limited.
The dataset can be download here.
- Classification
- Segmentation
- Classification failure cases:
- Segmentation on unlabelled images: