Skip to content

Latest commit

 

History

History
43 lines (34 loc) · 940 Bytes

model_training.md

File metadata and controls

43 lines (34 loc) · 940 Bytes

Model Training

more details in a dedicated model training page see full training scripts in training scripts folder. We use weights and biases for logging. Check the previous training runs here

Detailed Setup

Python Environment

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

we use weights and biases for logging. Make sure, you have an account and can login per console

pip install wandb
wandb login

Model Training

specify --download to download the data to the specified --data-path.

python train.py  \
  --data-path /data/marinedebris  \
  --model unet++  \
  --workers 32 \
  --batch-size 256  \
  --project marinedebrisdetector \
  --run-name unet++1 \
  --seed 1

Test Model

python test.py \
  --data-path /data/marinedebris \
  --ckpt-folder checkpoints/unet++1