Classifying-defected-parts

How to predict using the model:-

To predict on new image or folder of images, please provide path in --images_path and excute the below command.

python ./label_image.py \
--images_path={Path of the Image file or Folder of images} \
--trained_model_path = ./model_v7.h5 \
--incept_model_path = ./incept_model.h5 

An example on how to run this in a windows system with Anaconda :

C:/ProgramData/Anaconda3/envs/tensorflow/python.exe "c:/Users/Panda/Downloads/Upload_Github/label_image.py" --images_path=C:/Users/Panda/Downloads/Test --trained_model_path=C:/Users/Panda/Downloads/Upload_Github/model_v7.h5 --incept_model_path=C:/Users/Panda/Downloads/Upload_Github/incept_model.h5

The trained model achieved ~88 % accuracy with following parameters :-

• Epochs = 100
• Batch size = 32
• Learning Rate = 1e-5 (RMSprop)

Ref:-

1. https://www.cse.ust.hk/~qyang/Docs/2009/tkde_transfer_learning.pdf
2. https://towardsdatascience.com/a-comprehensive-hands-on-guide-to-transfer-learning-with-real-world-applications-in-deep-learning-212bf3b2f27a