A repository for collaborating on PANDA challenge.
- Solved tqdm and jupyter notebook dependency issue by（already added to dockerfile）
conda install -c conda-forge ipywidgets
pip install --upgrade jupyter_client- Solved skimage open tiff image issue by installing latest tiff package (already added to dockerfile)
pip install tifffileGenerate a set of tiles for a given slide.
Tile generation is computational expensive, so use pre-generated LMDB file for training.
from preprocessing.tile_generation import generate_grid
from preprocessing.normalization import reinhard_bg
test_slides_dir = "/PANDA_Challenge/train_images/"
image_id = "example"
im_size = 512 # Size of tile to be extracted at highest scanning magnification
overlap = 0.25 # Overlap percentage for tiles
ts_thres = 0.6 # Discard tiles with tissue region less that ts_thres
dw_rate = 4 # Tiles downsample rate
# Initialize a tile generator
tile_generator = generate_grid.TileGeneratorGrid(test_slides_dir,
f"{image_id}.tiff", masks_dir=None, verbose=False)
# Initialize a normalizer
tile_normalizer = reinhard_bg.ReinhardNormalizer()
# use the pre-computed LAB mean and std values
tile_normalizer.fit(None)
# Need to specify masks_dir for tile_generator if you want to generate labeled masks for each tile.
# Orig_tiles: tiles without normalization; norm_tiles: normalized tiles.
orig_tiles, norm_tiles, locations, tissue_masks, label_masks \
= tile_generator.extract_all_tiles(im_size, overlap, ts_thres, dw_rate, tile_normalizer)An example dataloader can be found in prediction_models/att_mil/datasets/test_slides.py