search

davidtvs / PyTorch-ENet

PyTorch implementation of ENet
MIT License
383 stars 129 forks source link

There are total 13 classes in camvid.py, but the outputs generated via pre-trained model in "./save/CamVid" have only 12 dimensions. #29

Closed veophi closed 4 years ago

veophi commented 4 years ago

The performance of the pre-trained model seems so poor. Thanks.

davidtvs commented 4 years ago

The road marking class is merged with the road class in the CamVid dataset (see main.py), so 12 classes are output.

Performance is similar to what the authors of ENet report in the paper - for CamVid this repo score 52.8 and the paper reports 51.3 class IoU. If you need better performance you need to go for a different NN.