https://github.com/fastmachinelearning/cloud-segmentation
https://github.com/fastmachinelearning/cloud-segmentation
Last synced: about 2 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/fastmachinelearning/cloud-segmentation
- Owner: fastmachinelearning
- Created: 2025-03-06T09:23:33.000Z (4 months ago)
- Default Branch: main
- Last Pushed: 2025-04-15T12:48:42.000Z (3 months ago)
- Last Synced: 2025-04-15T13:45:41.866Z (3 months ago)
- Language: Python
- Size: 535 KB
- Stars: 0
- Watchers: 15
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Package description
This folder contains example training scripts of AGENIUM SPACE tiny unet model on ALCD Cloud DB. It contains training script, the ALCD DB reprocessed in RGB a docker recipe.
# Docker
The folder contains a docker recipe and a make file to build an run it, got check the Readme in DOCKER folder. It is assumed you have a NVIDIA GPU on your computer.
To build the docker:
```sh
cd DOCKER
make build
```
To run the docker :
```sh
cd DOCKER
make bash
```
# Database
You can find the DB in DATA. It is a tiled and rgb format of the ALCD Cloud DB available online [here](https://zenodo.org/records/1460961).
# Trained Models
This section assume you are using the provided docker.
For training the tiny unet 100k model use :
```sh
python ./SCRIPTS/train.py run --model ags_tiny_unet_100k --data_path ./DATA/
```
For training the tiny unet 50k model use :
```sh
python ./SCRIPTS/train.py run --model ags_tiny_unet_50k --data_path ./DATA/
```
You can access the help info using
```sh
python ./SCRIPTS/train.py run -- --help
```
To use 2 gpus:
```bash
# using torchrun
torchrun --nproc_per_node=2 ./SCRIPTS/train.py run --model ags_tiny_unet_100k --data_path ./DATA/ --backend="nccl"
```
python ./SCRIPTS/train.py run --model ags_tiny_unet_50k --data_path ./DATA/
# Models Score
The table below contains the score results for the the models
| Model | Train Score (mean F1-score) | Valid Score (F1-score) |
|-------|-------------------|----------------|
| tiny_unet_50k | 0.94 - 0.78 | 0.94 - 0.77 |
| tiny_unet_100k | 0.94 - 0.79 | 0.84 - 0.78 |