Classifying single cells from the beehive into egg, larva, pollen and so on... I used single images solely from broodmapper.com.
- data/ pictures from broodmapper.com
- data_bayer segmented and labeld Bayer pictures
- preprocessing/ pre-processing, test-partitioning, and finding transformations
- dbs/ lmdbs of pre-processed and partitioned datasets
- major_classes/ I tried out a few settings. This one worked quite well. I excluded underrepresented classes from the training.
- deploy/ prepare model for deployment in REST service
- utils/ shared code
- label_maps.json label maps for broodmapper dataset
Note: This was one of my first projects using pytorch. There are a lot of things I unnecessarily did by hand. Nowadays, I would recommend using a library like pytorch-lightning.
Also: You might notice my use of lmdb. The data is not that big, but my PC was really bad at that time and I did not have a cloud available.