This repository contains an overview and short papers reviews of WSSS methods for histopathological image data. Occasionally, WSSS papers dealing exclusively with semantic segmentation on natural images (e.g. from PASCAL-VOC 2012 or ADE20K) will be included. I will update this list continuously during my current master thesis.
The ground-truth annotation type used for training is mentioned in the Annotation column for each paper. For WSSS, it generally falls into one of the following categories (in order of decreasing informativeness):
- Bounding box
- Scribble
- Point
- Image label
The LOD column roughly indicates the level of detail of the segmentation for each paper, e.g. whether a class is assigned to each pixel or only to each patch. This is a very vague information, since the pathology images have a different resolution for the different methods and the chosen patch size differs too. Therefore the LOD should only serve as a rough estimation for the time being.
Conference | Title | Annotation | LOD | Notes | Offical Code | Datasets |
---|---|---|---|---|---|---|
ICCV 2017 | Simple Does It: Weakly Supervised Instance and Semantic Segmentation | bounding box (multi-class) | pixel-level | coming soon | no | PASCAL VOC2012, VOC12+COCO |
Below is an overview of histopathology datasets from various body regions and tumor types. These can be used either for training or evaluation of classification or segmentation algorithms. The columns have the following meaning:
- #Labels/Img: Number of labels per image (BG = background)
- #Classes: How many different classes an image can be contain (most datasets only differentiate between tumor and background)
- #Img: Total number of images in the dataset
- #PGT: Number of images which have a pixel-wise ground-truth annotation (some images may only have image-level labels)
Name | Type | #Labels/Img | #Classes | #Img | #PGT | Image Size | Resolution | Paper |
---|---|---|---|---|---|---|---|---|
CAMELYON16 | Lymph node metastasis | 1 | 1 | 400 | 400 | 100k x 100k | 0.25 microns/px | yes |
CAMELYON17 | Lymph node metastasis | 1 | 1 | 1000 | 500 | 100k x 100k | 0.25 microns/px | yes |
PatchCamelyon | Lymph node metastasis | 1 | 1 | 327.680 | 0 | 96 x 96 | 0.25 microns/px | no |
CoCaHis | Colon cancer | 1 + BG | 1 + BG | 82 | 82 | 1037 x 1388 | 0.45 microns/px | yes |