search

TRI-ML / packnet-sfm

TRI-ML Monocular Depth Estimation Repository
https://tri-ml.github.io/packnet-sfm/
MIT License
1.23k stars 242 forks source link

"Re-training" your pretrained model #176

Open manavsingh415 opened 3 years ago

manavsingh415 commented 3 years ago

Hi. In order to train a packnet model on my custom dataset, I want to use your pretrained model PackNet01_MR_semisup_CStoK.ckpt, and further train this for a few epochs on my dataset.

I am hoping your pretrained model provides good initial weight values and will be able to fit my dataset with a few epochs. Whether or not this will work is a mystery.

Question is: How can I continue the training from a checkpoint, such that it starts taking input data from a different directory?

Thanks

EphChem commented 2 years ago

Have you figured this out @manavsingh415 ?

VitorGuizilini-TRI commented 2 years ago

The easiest way is to add the pre-trained model in the configuration file of a new training session, so the model loads the weights. Let me know if that works!