# Copy dataset/ into this directory# (Using python 3.8.13)
pip3 install -r requirements.txt
cd script/
# Use model.pth to perform prediction
python3 eval.py
# Train your own model
python3 main.py
Current results
model1.pth
Overall weighted IoU: 0.7649
Average true negative rate: 0.8996
Benchmark score: 0.8053
model4.pth
Overall weighted IoU: 0.7756
Average true negative rate: 0.8841
Benchmark score: 0.8082
model12.pth
Overall weighted IoU: 0.7745
Average true negative rate: 0.9966
Benchmark score: 0.8411
model14.pth
Overall weighted IoU: 0.9340
Average true negative rate: 0.9703
Benchmark score: 0.9449
TODO
Let the model predict confidence score
Deal with imbalance data (e.g. focal loss, dice loss, etc.)