I followed the tutorial for retraining an existing model, i successfully retrained a model for gender detection, but when i converted the .pb model to .tflite and used it on android using ML kit the accuracy went down, it still recognized male as female sometimes recognized male with 51%
@juniel03 on my opinion that could be because as far as you know that tflite format is a "compressed" version of the model. So some of the neutrons or weightage could be 'removed' to save the size.
I followed the tutorial for retraining an existing model, i successfully retrained a model for gender detection, but when i converted the .pb model to .tflite and used it on android using ML kit the accuracy went down, it still recognized male as female sometimes recognized male with 51%