Hi Team ,
I recently tried to use the Inference Engine of OpenVino for my pre-trained Image Classification model. My accuracy without using OpenVino is approximately 93% , but after using OpenVino it dropped to almost 80%.
Following were my observation, which could help you guys to understand the problem:
1) The accuracy for some classes of the model were really low, while other classes showed less incorrect predictions.
2) I use the resnet50, preprocess_input() for preprocessing the images in my actual model, which I was not able to use here.
Would be great to know of any possible solution to increase the accuracy.
Just to let you know, the above processing were all done on Intel Processor, 2.9 GHz Intel Core i7
Looking forward for any suggestions that would help me use the toolkit .