There is a newer version of the record available.

Published October 6, 2022 | Version v0.1.4
Software Open

okunator/cellseg_models.pytorch: v0.1.4

Authors/Creators

  • 1. University of Helsinki

Description

Test
  • Update loss tests
Fixes
  • Add a conv block BasicConvOld to enable Dippa to cellseg conversion of models.
  • Fix inst_key, aux_key bug in MultiTaskUnet

  • Add a type_map > 0 masking for the inst_maps in post-processing

  • Modify the optimizer adjustment utility function to adjust any optim/weight params.

  • Modify lit SegmentationExperiment according to new changes.

Features
  • Add optional spectral decoupliing to all losses
  • Add optional Label smoothing to all losses

  • Add optional Spatially varying label smoothing to all losses

  • Add mse, ssim and iqi torchmetrics for metric logging.

  • Add wandb per class metric callback for logging.
  • Add from_yaml init classmethod to initialize from yaml files.

Files

okunator/cellseg_models.pytorch-v0.1.4.zip

Files (8.6 MB)

Name Size Download all
md5:de1a9b0ed89fbcb15b19d8bde4f6466c
8.6 MB Preview Download

Additional details

Related works

Is supplement to
https://github.com/okunator/cellseg_models.pytorch/tree/v0.1.4 (URL)