WUT-AI-AI4Mat/microstructure-micro-sam

MicroSAM Fine-Tuned Checkpoints for Microstructure Segmentation

Model Description

This repository contains dataset-specific fine-tuned parameter checkpoints for the MicroSAM vit_b_lm model. The files contain fine-tuned parameters rather than a complete base model. The matching scripts obtain the vit_b_lm base model through micro_sam.util.get_sam_model and then load these parameters.

Source code: https://github.com/WUT-AI-AI4Mat/Segmentation-methods-evaluation-for-quantitative-microstructure-analysis

Fine-Tuning Routes and Checkpoints

Dataset	Route	Classes	File
Aachen-Heerlen	LoRA plus mask decoder	2	checkpoints/Aachen-Heerlen/microsam_lora_decoder_best.pth
EMPS	LoRA plus mask decoder	2	checkpoints/EMPS/microsam_lora_decoder_best.pth
Grain	LoRA plus mask decoder	2	checkpoints/Grain/microsam_lora_decoder_best.pth
EBC	LoRA, mask decoder, and class tokens	3	checkpoints/EBC/semantic_microsam_best.pth
Super	LoRA, mask decoder, and class tokens	3	checkpoints/Super/semantic_microsam_best.pth
MetalDAM	LoRA, mask decoder, and class tokens	5	checkpoints/MetalDAM/semantic_microsam_best.pth
UHCS	LoRA, mask decoder, and class tokens	7	checkpoints/UHCS/semantic_microsam_best.pth

Class counts include background.

Base Model Requirement

Initialize the required base model with:

python -c "from micro_sam.util import get_sam_model; get_sam_model(model_type='vit_b_lm')"

The corresponding BioImage.IO checkpoint is available at: https://uk1s3.embassy.ebi.ac.uk/public-datasets/bioimage.io/diplomatic-bug/1.2/files/vit_b.pt

Training Details

• Base architecture: MicroSAM ViT-B-LM
• SAM encoder input: longest side resized and padded to 1024 x 1024
• Training batch size: 1
• Epochs: 200
• Optimizer: AdamW
• Learning rate: 0.0001
• Weight decay: 0.0001
• Scheduler: CosineAnnealingLR with minimum learning rate 0.000001
• Early-stopping patience: 50
• LoRA rank: 8
• LoRA alpha: 16
• LoRA dropout: 0.05
• LoRA targets: qkv and proj
• Binary objective: BCE, Dice, and IoU MSE
• Multiclass objective: cross-entropy and Dice

Usage

Example for binary EMPS segmentation:

hf download NAMESPACE/microstructure-micro-sam checkpoints/EMPS/microsam_lora_decoder_best.pth --local-dir weights/micro-sam
python microsam/test_lora_decoder.py \
  --dataset-root /path/to/EMPS \
  --finetuned-checkpoint weights/micro-sam/checkpoints/EMPS/microsam_lora_decoder_best.pth \
  --output-dir results/micro_sam_emps

For multiclass checkpoints, use microsam/test_semantic_microsam.py and pass the class count from the table.

Inference Parameters

Automatic instance segmentation uses min_size=10, center_distance_threshold=0.5, and boundary_distance_threshold=0.5.

Evaluation, Intended Use, and Limitations

Evaluation saves original-resolution masks and calculates the common benchmark metrics through Myutils/metrics.py. These checkpoints require the matching MicroSAM vit_b_lm base model and source environment. MicroSAM dependencies such as vigra and python-elf are most reliably installed through the official Conda environment on Linux. The checkpoints are intended for research reproduction and are not validated for safety-critical or industrial quality-control decisions.

License and Citation

The released experiment files are provided under the MIT license. MicroSAM and its base model remain subject to upstream terms and the base checkpoint is not redistributed here. A paper citation will be added after publication.