---
license: mit
library_name: mmsegmentation
pipeline_tag: image-segmentation
tags:
- semantic-segmentation
- microstructure
- materials-science
- pytorch
- segformer
- mmsegmentation
---

# SegFormer-B0 for Quantitative Microstructure Segmentation

## Model Description

This repository contains seven dataset-specific SegFormer checkpoints trained
with MMSegmentation. The architecture uses a MiT-B0 backbone and SegFormer
decode head with 512 x 512 inputs.

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

## Checkpoints

| Dataset | Classes | File |
| --- | ---: | --- |
| Aachen-Heerlen | 2 | `checkpoints/Aachen-Heerlen/best_mIoU_epoch_76.pth` |
| EMPS | 2 | `checkpoints/EMPS/best_mIoU_epoch_153.pth` |
| Grain | 2 | `checkpoints/Grain/best_mIoU_epoch_63.pth` |
| EBC | 3 | `checkpoints/EBC/best_mIoU_epoch_128.pth` |
| Super | 3 | `checkpoints/Super/best_mIoU_epoch_170.pth` |
| MetalDAM | 5 | `checkpoints/MetalDAM/best_mIoU_epoch_96.pth` |
| UHCS | 7 | `checkpoints/UHCS/best_mIoU_epoch_137.pth` |

Class counts include background.

## Training Details

- Backbone: MiT-B0
- Input size: 512 x 512
- Training batch size: 32
- Validation and test batch size: 16
- Epochs: 200
- Optimizer: AdamW
- Learning rate: 0.00006
- Adam betas: 0.9 and 0.999
- Weight decay: 0.01
- Scheduler: five-epoch linear warm-up followed by polynomial decay
- Initialization: `mit_b0_20220624-7e0fe6dd.pth`

## Usage

Install the MMSegmentation environment described in the source repository.
Example for EBC:

```bash
hf download NAMESPACE/microstructure-segformer-b0 checkpoints/EBC/best_mIoU_epoch_128.pth --local-dir weights/segformer-b0
cd mmsegmentation
export DATASET_ROOT=/path/to/EBC
python tools/test_ebc_segformer.py \
  --dataset-root /path/to/EBC \
  --checkpoint ../weights/segformer-b0/checkpoints/EBC/best_mIoU_epoch_128.pth \
  --result-root ../results/segformer_ebc
```

Use the corresponding dataset config and `tools/test_*_segformer.py` wrapper
for the other checkpoints.

## Evaluation

The shared testing code calls `Myutils.metrics.Metric.compute_all`, saves raw
prediction masks, and exports an Excel report. Reported metrics include mIoU,
Dice, precision, recall, accuracy, HD95, Hausdorff distance, NSD, MAE, MBSS,
and MBSS_add. Numerical benchmark results will be linked after the associated
paper becomes publicly available.

## Intended Use and Limitations

The checkpoints are intended for reproducing segmentation experiments on the
named datasets. Performance may not transfer to unseen materials, imaging
modalities, magnifications, or annotation conventions. The models are not
validated for safety-critical or industrial quality-control decisions.

## License and Citation

The released experiment files are provided under the MIT license.
MMSegmentation and the SegFormer initialization remain subject to their
upstream licenses. A paper citation will be added after publication.