Issues with training
stopcharlaCapillary opened this issue · comments
Hey, I have tight bounded face images, the sizes vary from(80x80 to 220x220). I am trying to fine tune inception v3. I have trained with around 8000 images. The data is not that biased too. But once I try to test it out the output is either male or all the images in folder or female for all images with probability of 1. I'm not sure what mistake I'm doing here. Can you please help me with this.
I'm using your prepoc.py , train.py and guess.py as is.
I gave the labels as 0 and 1 in train.txt
Cannot replicate. Would need more information about your data. Works fine for me