Added custom feature extractor model and a preprocessing function for that model

[ibadrather]

A very good and useful repository.

I wanted to use my custom model I trained on a specific dataset to use for feautre extraction for the Reverse Image Search App I am working on.

Deep Image Search library provided everything I needed except the possibility of using my own model. So, I added that functionality to your Library. All the old functionality of the library is there but now you can use your own pretrained model in any format like onnx, torchscript, even tensorflow, or keras or scikit-learn models.

I keep did this by making small changes in selecting the model used for feature extraction. I did not do anything that will increase the dependence on other repos. I will write an example code if you are okay with me to work on this repository with you.

In future I am thinking of adding dimenionality reduction feature for this along with other few changes.

On my side I made sure that the code is backwards compatible. If you find any issues, please let me know.

If we continue to work on this repository together, we can added automated testing to this to make sure it is backwards compatible everything.

If you have any questions or any suggestions, I am open to hearing them.

I also want to use this package in my server, so, I would appreciate if you can publish the changes on PyPi if you deem them to be correct.