search

VITA-Group / FasterSeg

[ICLR 2020] "FasterSeg: Searching for Faster Real-time Semantic Segmentation" by Wuyang Chen, Xinyu Gong, Xianming Liu, Qian Zhang, Yuan Li, Zhangyang Wang
MIT License
526 stars 107 forks source link

Could I use the search result to train model directly ? #8

Closed nieyan closed 4 years ago

nieyan commented 4 years ago

Thank you for your great work!!! Search process may take a long time to run. Want use the citysacpe dataset search result model to train my own data at first. But I can't find that code.

chenwydj commented 4 years ago

Thank you for your interest in our work!

  1. You need to provide the dataloader for your data (see examples here).
  2. The current provided training configuration file is already using the searched architecture from Cityscapes. So you can use the provided training code to train on your data with our searched model.
  3. One thing you may need to be aware of is that your num_classes equals to that of Cityscapes which is 19.