search

hugochan / Eye-Tracker

Implemented and improved the iTracker model proposed in the paper "Eye Tracking for Everyone"
BSD 3-Clause "New" or "Revised" License
232 stars 52 forks source link

Can I have more information about smaller dataset? #4

Closed leconize closed 3 years ago

leconize commented 6 years ago

Hi Hugo. I see you have mentioned the small dataset in the readme. Can you provide me more information about the datasets? I would like to know how you obtain it.

hugochan commented 6 years ago

The original dataset comes from the GazeCapture project. The dataset involves over 1400 subjects and results in more than 2 million face images. Due to the limitation of computation power, a much smaller dataset with 48000 training samples and 5000 validation samples was used here. Each sample contains 5 items: face, left eye, right eye, face mask and labels.

rofailag18 commented 6 years ago

Chan, I have three questions: (1) thank you for providing this nice dataset, but how we would like to separate the validation dataset from the training dataset. It looks like the file you provided is numpy data file. I have to load it using numpy and then save the validation set as a separate data file when using your validation script. Also, your validation script validation_script.py requires data file (which I assume this is) and the model file (which you provided as model-23.data-00000-of-00001).

(2) Once we have the validation data and the model file, I can use these two files as the inputs for validation_script.py, right?

(3) How would your suggest me to test your model with an face image?

Thank you

hugochan commented 6 years ago

@rofailag18 For running the validation script, just run python validation_script.py pretrained_models/itracker_adv/model-23 -v eye_tracker_train_and_val.npz.

You might want to split the dataset further into train, validation and test set for fair experiments.

FuHongy commented 5 years ago

@hugochan Hi! I'm wondering that does it need an extra test set when testing the pretrained model?

hugochan commented 5 years ago

@FuHongy Yes, it needs. You might want to further split the current training set into training and dev sets and use the current dev set as a test set.