search

TRI-ML / packnet-sfm

TRI-ML Monocular Depth Estimation Repository
https://tri-ml.github.io/packnet-sfm/
MIT License
1.24k stars 243 forks source link

Training from the pretrained model #69

Closed tonytu16 closed 4 years ago

tonytu16 commented 4 years ago

Hello,

I am trying to train from PackNet01_HR_velsup_CStoK.ckpt using my own dataset. So my training command is python train.py PackNet01_HR_velsup_CStoK.ckpt However I get a FileNotFoundError: [Errno 2] No such file or directory: '/data/datasets/KITTI_raw/data_splits/eigen_test_files.txt' error

It looks like the model is still expecting the KITTI dataset. Is there a way to change this? Thank you!

VitorGuizilini-TRI commented 4 years ago

If you provide the checkpoint like that the training will resume, will presumes that the same dataset is available. If you want to fine-tune a pre-trained model on your own dataset, you have to provide a new config (.yaml) file and include:

model:
    checkpoint_path: /path/to/file.ckpt

Let me know if that works!

tonytu16 commented 4 years ago

Thank you! Will this overwrite the pretrained model or will the fine-tuned version of the model be saved to a new folder?

VitorGuizilini-TRI commented 4 years ago

It will be saved to a new folder (remember to specify the details of where the fine-tuned model will saved in the config file).