search

czero69 / acomoeye-NN

Acomoeye-NN: NVGaze gaze estimation with upgrades
Apache License 2.0
23 stars 6 forks source link

Dividing NVGaze dataset for training and testing #3

Closed rsingh76 closed 2 years ago

rsingh76 commented 2 years ago

I am trying to follow this page and write a model in PyTorch. However, I am not sure about how to divide the NVGaze dataset for training/validation/testing. Could you give some insight into how you did it for both real and synthetic datasets?

  1. Do you already have the divided txt/csv that I can use?
  2. How long does it usually take to train on acomo dataset and NVGaze dataset?
  3. I am new to the field and using PyTorch, is there anything I should keep in mind while porting this to PyTorch?
czero69 commented 2 years ago

Sorry for replaying so late.

  1. For AcomoEye train/test split text files are included and a direct comparison can be made. For NVGaze you must create your own split (I want to avoid publishing Nvidia files here). You can follow the way how I did the train/test split.
  2. Few hours to one day.
  3. I think it should be quite straightforward (and good idea) to implement it in PyTorch