Learning Pytorch: From Zero to GANs

Original article was published on Deep Learning on Medium


I was shocked when I saw that, so I decided to test it using the test dataset available. Unfortunately, only 33 images were available for testing in test folder. So I decided to test them all and surprisingly, model was able to predict all the images correctly. Some for the predictions are given below.

So how is this model able to perform this nice? Actually, I don’t know. And I seriously doubt that validation set contains images from training dataset and and test dataset also contains same same images. If this is the case, then model must be learning the images which it was going to predict. Why do I think so? Because I had ran this code once before finalizing it and talked about it in the forum. It was seen that, I had kept images from training dataset in both validation dataset and training dataset and had achieved similar accuracy. But, in that case, model was able to predict 30 images correctly out of available 33. There could be other reasons for this much accuracy such as mistake in my code( which I had not found yet. Do tell if you are able to find it.), availability of large number of high resolution images and use of several improvement techniques.

I was planning to try out other architectures such as AlexNet, but later did not try because I was overwhelmed by the accuracy that I got by using ResNet9. Maybe in future I will try. The structure of the code is similar to the structure of the code used for teaching. So this is all about my work. You can check this notebook at Jovian(https://jovian.ml/omnarayant89/plant-disease-classification) or at Kaggle(https://www.kaggle.com/omnarayansharmalohar/plant-diseases-classification).

So now, what are other learnings from this course?

This is the first ever online course that I attended and I felt good. All the support I got from the forum is overwhelming. Though, in the middle I felt like I will not be able to do it as our teachers were adding numbers of assignments each day and content of the course was going over my head, I finally completed it and I am happy about it.

Hitherto I believed that one can learn a lot by being alone but, now I know that learning can be made quick and process can be made happier by having an amazing community for it.

Now I also know that, AI is not some obscure programming related pseudo-science, but it is based on use of mathematics in an amazing way.

One more thing, this also taught me that all the problem or the hurdles that exist between us and learning merely exist in our mind only. And internet is the ocean of knowledge we just need to have a will and we can learn whatever we want to. So good luck everyone. Happy Learning!