Releases: DIAGNijmegen/unicorn_baseline
Releases · DIAGNijmegen/unicorn_baseline
v1.4.4
- update patch size for task 6 and task 7 (#51)
- updated example patch-level neural representation to match new schema (#52)
- bugfix in radiology dense encoding (#52)
v1.4.3
- fix caching issue and improve preprocessing for language tasks (#49)
v1.4.2
- update task 6 patch size (#46)
- revert Virchow encoding to output class tokens for pathology dense prediction tasks (#47)
- increase max number of tiles from 14k to 30k in pathology vision tasks (#47)
v1.4.1
- add patch-token support (#43)
- optimize write_json_file with orjson (#43)
v1.4.0
- fix encoding error for task 16 (#39)
- fix encoding error for task 18 (#39)
- replace TITAN with PRISM for case-level pathology vision tasks (#40)
- replace CONCH v1.5 with Virchow for dense pathology vision tasks (#40)
v1.3.2
- explicitly add
xformers
version (#36)
v1.3.1
- fix path when loading
mrsegmentator
weights (#29)
- bugfix for the new format of task19 (#32)
1.2.2
- fix model path for radiology vision tasks
v1.2.1
- set
image_path
correctly in src/unicorn_baseline/vision/radiology/main.py
- clean up & black formatting