search

hysts / pytorch_mpiigaze

An unofficial PyTorch implementation of MPIIGaze and MPIIFaceGaze
MIT License
346 stars 85 forks source link

Some questions about mpiigaze dataset. #35

Closed Kelly-ZH closed 3 years ago

Kelly-ZH commented 3 years ago

Hello, hysts. I have two questions as below, and I am looking forward to your reply. Thank you very much.

  1. Why assert the length of images, poses, gazes are 3000 here ？ Is it because random select 1500 left eye images and 1500 right eye images from the full dataset?
  2. When I run

    python train.py --config configs/mpiigaze/lenet_train.yaml

why does step(len(train_loader)) is 1181? How to calculate this number？

hysts commented 3 years ago

@Kelly-ZH

  1. Yes.
  2. 3000 (number of images per person) 14 (number of person used for training) 0.9 (ratio of training data) / 32 (batch size). The ratio of validation data is specified here and the ratio of training data is 1 - val_ratio.
Kelly-ZH commented 3 years ago

You really did me a big favor. I know it. Thank you very much.