Commits

Branch selector

 main

User selector

All users

All time

Commit History

Commits on Jan 28, 2026

• fix typo

  moritzschramm

  committed
  8f6e436

  Copy full SHA for 8f6e436

• add future works section in readme

  moritzschramm

  committed
  d6d3a09

  Copy full SHA for d6d3a09

Commits on Jan 26, 2026

• make bash default shell in build job

  moritzschramm

  committed
  79d14ef

  Copy full SHA for 79d14ef

Commits on Jan 24, 2026

• add badge in readme

  moritzschramm

  committed
  cb07fc3

  Copy full SHA for cb07fc3

• fix build.yml 2

  moritzschramm

  committed
  1478230

  Copy full SHA for 1478230

• fix build.yml

  moritzschramm

  committed
  658820d

  Copy full SHA for 658820d

• fix setup script for github workflow

  moritzschramm

  committed
  c9211fa

  Copy full SHA for c9211fa

• fix directory name

  moritzschramm

  committed
  e2c95df

  Copy full SHA for e2c95df

• create github build workflow

  moritzschramm

  committed
  b5bee8d

  Copy full SHA for b5bee8d

• update readme

  moritzschramm

  committed
  1ac0268

  Copy full SHA for 1ac0268

• .hpp -> .h + .cpp

  moritzschramm

  committed
  24736b6

  Copy full SHA for 24736b6

Commits on Jan 23, 2026

• make canny_edge consistent with other edge cost algorithms; add switch to easily exchange edge cost algorithm; add watershed to compute edge costs; output more statistics for validation and improv…

  Show description for f185640
  moritzschramm

  committed
  f185640

  Copy full SHA for f185640

• create graph based edge costs; fix slic edge cost

  moritzschramm

  committed
  49fe5f3

  Copy full SHA for 49fe5f3

• ensure consistency of metrics and comments for model training

  moritzschramm

  committed
  aa880b4

  Copy full SHA for aa880b4

• fix image slicer filename generation, adapt parameters for training

  moritzschramm

  committed
  1092900

  Copy full SHA for 1092900

• expand model to 3 up and down layers; add precision, recall and f1 stats

  moritzschramm

  committed
  ffade88

  Copy full SHA for ffade88

Commits on Jan 21, 2026

• use BCE with logits as loss in pretraining for costs; use self-supervised calibration loss for sigma

  moritzschramm

  committed
  1049fca

  Copy full SHA for 1049fca

• fix loss in pretraining, change penalty in RL to undersegmentation

  moritzschramm

  committed
  9106bcb

  Copy full SHA for 9106bcb

Commits on Jan 20, 2026

• new loss to make sure negative sign in costs is not lost

  moritzschramm

  committed
  36daab3

  Copy full SHA for 36daab3

• fix compress.cpp

  moritzschramm

  committed
  bd171e7

  Copy full SHA for bd171e7

Commits on Jan 19, 2026

• fix pretraining and training

  moritzschramm

  committed
  19b14c8

  Copy full SHA for 19b14c8

Commits on Jan 17, 2026

• add evaluation to RL training

  moritzschramm

  committed
  d81dea8

  Copy full SHA for d81dea8

• improve performance of RL by 'batching' call to RAMA and compute_rewards

  moritzschramm

  committed
  cd543e9

  Copy full SHA for cd543e9

• refactor EdgeDataset

  moritzschramm

  committed
  5e63865

  Copy full SHA for 5e63865

• add validation to pretraining loop

  moritzschramm

  committed
  72c4acd

  Copy full SHA for 72c4acd

• improve pretraining performance

  moritzschramm

  committed
  795dcbe

  Copy full SHA for 795dcbe

• implement compression with segmentation from network inference

  moritzschramm

  committed
  4874de7

  Copy full SHA for 4874de7

• fix data type when accessing edge costs

  moritzschramm

  committed
  ae03e7a

  Copy full SHA for ae03e7a

• add target edge cost generation with SLIC superpixels

  moritzschramm

  committed
  a430caf

  Copy full SHA for a430caf

Commits on Jan 16, 2026

• fix target creation in pretraining; fix segment_extract in training

  moritzschramm

  committed
  48d0263

  Copy full SHA for 48d0263

Commits on Jan 15, 2026

• add canny edge detection to create pre-training targets

  moritzschramm

  committed
  a7ad0eb

  Copy full SHA for a7ad0eb

Commits on Jan 14, 2026

• create random partition as targets for pre-training

  moritzschramm

  committed
  d85126d

  Copy full SHA for d85126d

• implement compute_rewards

  moritzschramm

  committed
  7411cbb

  Copy full SHA for 7411cbb

Commits on Jan 7, 2026

• use 3 channels for training; added mask for pretraining; fixed bugs in compute_rewards and segment_extract

  moritzschramm

  committed
  626f850

  Copy full SHA for 626f850

Commits on Jan 1, 2026

• update readme

  moritzschramm

  committed
  e7b0cf7

  Copy full SHA for e7b0cf7