search

hszhao / ICNet

ICNet for Real-Time Semantic Segmentation on High-Resolution Images, ECCV2018
https://hszhao.github.io/projects/icnet
605 stars 199 forks source link

Training parameters set #24

Open InstantWindy opened 6 years ago

InstantWindy commented 6 years ago

Hello ! could you porvide me the training parameters set? beacause I train the model ,I found it can not reach the paper performance . Thank you ！

oandrienko commented 6 years ago

To train the model properly, you either need to do transfer learning from ResNet50 or PSPNet50 weights as suggested by the author. You also need to apply network compression once you have trained the full scale model.

If you are interested in a Tensorflow implementation, I have some pretrained models with the accompanying sets of hyperparameters here: https://github.com/oandrienko/fast-semantic-segmentation. Hope this helps!

pewpewpeww commented 5 years ago

@InstantWindy Hi, I am looking at training ICNet with my own data set too. Can you share your training prototxt for my reference? Thank you :-)