Added predction for folder of images

[kharyal]

The prediction method now works for both, single image and a folder of images. It can be quite a tiresome process to run the prediction for many images to get the general idea of how the network is performing on some custom dataset. Therefore, this pull request.

I have changed the following files:

• README.md

• predict.py

In Readme.md I have just changed the line that says prediction runs for a single image.

In predict.py I have added a few lines to handle the prediction on a folder of images. The code now does the following things along with the things that it did earlier:

• checks if ./predict is already available. If not, it makes ./predict folder. If it is available, deletes all the files (probably from previous runs) that are already there in the folder.

• Saves the pred_seg, blend_pred, mask, depth images as the input_file_name_<type>.png in ./predict folder where <type> is one of seg, blend, mask, depth.

Also, I want to thank you for such great work. I am sorry for any mistakes that I might have made in this PR because it is my first one ever.

[bertjiazheng]

Hi,

First of all, thanks for your PR!

Since this customized feature would not affect any result, I prefer not to merge this PR into the current master.

Thanks anyway.

Best, Jia

[kharyal]

Yes, sure. I just wanted to run the prediction for a large number of images, just to see how it performs on those images and therefore I changed the code. The PR was just a byproduct of that.

Thanks