Skip to content

Releases: clemsgrs/slide2vec

v2.0.1

08 Jan 12:32

Choose a tag to compare

  • sync hs2p to add support for background removal when encoding tiles (#46) and for HSV tissue seg (#48)
  • improve support for using locally trained DINO models (#49)

v2.0.0

16 Dec 13:16

Choose a tag to compare

  • replace preprocessing code with hs2p repo to avoid duplication and ease maintenance (#42)

v1.4.0

04 Dec 16:35

Choose a tag to compare

  • add CONCH, kaiko, MUSK, PhikonV2 and Hibou tile-level FMs (#37)
  • allow saving tile-level embeddings along side slide-level embeddings for slide-level FMs (#35)
  • allow saving PRISM latents (#36)
  • add kaiko's Midnight12k tile-level FM (#39)

v1.3.0

21 Oct 15:00

Choose a tag to compare

  • merged package dependencies in a single requirement.in file (#31)
  • added PR test to ensure consistency in tiling & embedding (with PRISM) script (#32)

v1.2.3

05 Sep 12:24

Choose a tag to compare

  • restored deduplication logic to prevent duplicate tile features when number of tiles is not divisible by number of gpus (#29)

v1.2.2

07 Aug 09:32

Choose a tag to compare

  • tiny round fix when loading and rescaling tissue mask (#26)
  • improved logging when loading weights from local path (#26)
  • improved safety when resume: true (#26)

v1.2.1

22 Jul 21:03

Choose a tag to compare

  • updated ASAP to 2.2 (#24)
  • fixed some issues when reading coordinates from an existing folder
  • added H0-mini to list of supported tile encoder (#24)
  • added code for feature extraction with our in-house DINO / DINOv2 models (#24)

v1.2.0

30 May 10:54

Choose a tag to compare

  • fix SIGBUS error by moving from numpy mmap file to h5py datasets (#22)

v1.1.2

22 May 22:50

Choose a tag to compare

  • fixed visualization bug due to wrong handling of tile_size_at_0
  • fixed gpu oom / cpu oom errors when encoding large whole slides into tile-level features

v1.1.1

15 May 10:25

Choose a tag to compare

  • remove need for common spacings when loading tissue mask
  • improve gpu mem management
  • improved distributed flexibility with fetching free port