I am trying to create an android app that classifies an image. The image classification model was first produced using Python and Keras. The accuracy of the test data set was approximately 93%. I converted the model file, which was saved to the filesystem as an h5 file to tflite by following the tensorflow documentation. The conversion included Quantization.
However, after integrating the Model to the app, I always see the same 3 to 4 labels as the result to whatever the image I use as the input.
I see a big failure here. Anybody finds any problem with this approach, please point me in right direction.