Adaptive Polishing: SEM training data & machine-learning segmentation models

Datasets of SEM images of ceullular cryo-FIB lamellae during the polishing stage. This repository contains all the training data and models used for developing and validating Adaptive Polishing in the manuscript: Adaptive Milling: "imaging feedback-driven automated fabrication of consistently thin cryo-lamellae". Trained models Models trained with AM-model-training (https://github.com/rosalindfranklininstitute/AM-model-training) are provided, together with CSV files containing the relative paths to image label pairs used for model training and validation. Model weights file Training .csv file Validation .csv file description AM_SEM_Arctis_V01.pth Franklin_train_data.csv Franklin_val_data.csv Original model trained with Arctis data only AM_SEM_All_V01.pth all_train_data.csv all_val_data.csv Model trained with all the data. --> use on microscope AM_SEM_Hydra_V01.pth hydra_train_data.csv hydra_val_data.csv Refined from Arctis model with Hydra data AM_SEM_Aquilos2_V01.pth aquilos2_train_data.csv aquilos2_val_data.csv Refined from Arctis model with Aquilos 2 data AM_SEM_Hydra_Aquilos2_V01.pth hydra_aquilos2_train_data.csv hydra_aquilos2_val_data.csv Refined from Arctis model with Hydra and Aquilos 2 data Training data 3186 SEM images of cryo-FIB lamellae with corresponding labels, used to train machine-learning segmentation models for Adaptive Polishing (https://github.com/rosalindfranklininstitute/adaptive_milling). The data was acquired in 20 microscope sessions, using either a Thermo Fisher Arctis, Aquilos 2 or Helios Hydra FIB/SEM microscope. Biological and microscope metadata are provided per SEM images in metadata.csv. SEM images and labels are in 8 or 16-bit .tif format with dimensions of 3072 x 2048 pixels. sessions overview session_id session_name Microscope sample lamella_type session_001 2025_eBIC Aquilos 2 mESC cells-on-grid session_002 20230830 Arctis HeLa cells-on-grid session_003 20231102 Arctis RPE-1 Tet-On PLK4 GFP-centrin cells-on-grid session_004 20240126_Adaptive_milling Arctis RPE-1 Tet-On PLK4 GFP-centrin cells-on-grid session_005 20240223_adaptive_milling_ai Arctis RPE-1 Tet-On PLK4 GFP-centrin cells-on-grid session_006 20240228_adaptive_milling_ai Arctis RPE-1 Tet-On PLK4 GFP-centrin cells-on-grid session_007 20240321_CElegans_Adaptive_milling Arctis Caenorhabditis elegans lift-out session_008 20240624_Adaptive_milling Arctis RPE-1 Tet-On PLK4 GFP-centrin cells-on-grid session_009 20240628_adaptive_autolamella Arctis RPE-1 Tet-On PLK4 GFP-centrin cells-on-grid session_010 20240705_adaptive_milling Arctis RPE-1 Tet-On PLK4 GFP-centrin cells-on-grid session_011 20240806_auNP Arctis HeLa cells-on-grid session_012 20240916_adaptive_waffleyeast Arctis Yeast Waffle session_013 20241007_adaptive_vs_non-apadaptive_01 Arctis RPE-1 Tet-On PLK4 GFP-centrin cells-on-grid session_014 20250414_adaptive_milling Arctis mESC cells-on-grid session_015 20250428_AM Arctis mESC cells-on-grid session_016 20250513_AM_ON Arctis mESC cells-on-grid session_017 20250519_AM_troubleshoot Arctis mESC cells-on-grid session_018 20250528_AM_vs_nonAM_01 Arctis mESC cells-on-grid session_019 20250610_quick_ON_test Arctis mESC cells-on-grid session_020 20250905_rosalind Helios Hydra RPE-1 Tet-On PLK4 GFP-centrin cells-on-grid Annotation labels: Pixel Value Label 0 Background 1 GIS layer 2 Lamella 3 Cracks and holes 4 Vacuum