Skip to content

v1.0: pre-trained models

Latest

Choose a tag to compare

@charlesnchr charlesnchr released this 15 Nov 13:23
· 10 commits to master since this release
1befa53

Pre-trained models for ML-SIM and the desktop app.

File Model SHA-256
DIV2K_randomised_3x3_20200317.pth ML-SIM reconstruction model from the paper. RCAN with 9 input channels, 3 residual groups, 10 blocks per group and 96 features, for stacks of 3 orientations and 3 phase shifts. e4936f5ccf5db42009fa23ca3cbd63f53125dbd787240ca78f09e2b85c682a08
meng_3colours_4.pth ERnet segmentation model (background, sheet, tubule) loaded by the ERnet plugin of the desktop app. 5b229eb9775654affa86e6c8b959afeb5d3ca0c76e4c8b6a8096c738e2456620
20210703-ERnet-4class-3kSamples.pth Later four-class ERnet model. Not used by the code in this repository, kept from the old download server. 60f59b9b1acc3cf2eacb61272ee16c4d85db12ebcc4f8d1277d9b1a52a0be1b5

python reconstruct.py Test_data/*.tif downloads the ML-SIM model on first use, and the desktop app downloads the first two. These files replace the old S3 bucket, which is no longer available.