search

icatcherplus / icatcher_plus

iCatcher+: Robust and automated annotation of infant gaze from videos collected in laboratory, field, and online studies
GNU General Public License v3.0
17 stars 17 forks source link

Fix hard coded values #59

Closed AminSarafraz closed 1 year ago

AminSarafraz commented 1 year ago

Based on our discussion, the following values are hard coded, fixing it will help with training using other datasets

reproduce/models.py line 314
    self.fc3 = torch.nn.Linear(512, 3)

reproduce/preprocess.py line 467
    classes = {"away": 0, "left": 1, "right": 2}

reproduce/preprocess.py line 625
    classes = {"away": 0, "left": 1, "right": 2}
yoterel commented 1 year ago

For now, I decided to leave this untouched as the reproduce folder is used to replicate results rather than allowing extensions. Anyone wanting to retrain will need to manually step through the code anyway, and this comment might serve as a hint as to what needs changing for a 4 class (or more) training.