# segmentation

**Repository Path**: MoLongCode/segmentation

## Basic Information

- **Project Name**: segmentation
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2024-12-16
- **Last Updated**: 2025-01-03

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# simple-unet-2d
simple unet with NeurIPS'19 topoloss

**Commands:**

* Make sure to populate `train.json` and `test.json` with appropriate hyprerparameters

**Train:**
```
CUDA_VISIBLE_DEVICES=3 python3 main.py --params ./datalists/DRIVE/train.json
```
* Ensure `crop_size` in `train.json` is divisible by 16

**Test/Inference:**
```
CUDA_VISIBLE_DEVICES=4 python3 main.py --params ./datalists/DRIVE/test.json
```
**Compute Evaluation Metrics (Quantitative Results):**
```
python3 compute-eval-metrics.py
```
**Dataset properties:**

GT: Foreground should be 255 ; Background should be 0

* First do pretrain (1000-2000 epochs) by setting `"topo_weight": 0` in `train.json`
* Then, rainload the best model from pretrain and t using topoloss by setting `topo_weight` to a non-zero value. Change the `output_folder` and `checkpoint_restore` in `train.json` too