Issue Training Gender

[osseney]

Even after training gender for 30k steps at 0.001, guess.py only guesses ONE value.

For example, it will guess the gender of the first picture to be F at 0.56 probability. It will also guess every other picture I feed it as F at 0.56 probability, even if the picture should be M.

Here's the command I use:

python train.py model_type bn --train_dir /home/autumn/tensor/AgeGenderDeepLearning-master/Folds/gender_test_fold_is_0 --max_steps 30000 --eta 0.001

python guess.py --model_dir ~/tensor/AgeGenderDeepLearning-master/Folds/gender_test_fold_is_0/run-142133/ --class_type gender --filename ~/tensor/cnnimages/

Works fine for age, but not for gender.

[dpressel]

The docs for the learning rate and LRS were specifically for the original Levi-Hassner model (see #35). You might have to experiment a bit to get a learning regimen that works for that model

[osseney]

The issue was the checkpoints for the folds. I have to delete the fold every time I use it and preprocess it again or it refuses to train itself for some reason. Thanks On Monday, June 22, 2020, 8:01:09 PM MDT, Daniel Pressel <[email protected]> wrote: The docs for the learning rate and LRS were specifically for the original Levi-Hassner model (see #35). You might have to experiment a bit to get a learning regimen that works for that model — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.