Inference on a directory of images

[LukasBommes]

It would be handy if the inference script could handle not only a single image, but an entire folder of images. Since looking at only a single example does not provide a complete picture of the algorithm's performance.

I know, this could be achieved with a bash script. But it is easier for newcomers to have this functionality built into the inference script.

I would automatically disable graphical output (imshow) when performing inference on an entire folder (requring specifiction of the save_path in that case). The directory and its subdirectories could be recursively traversed and the outputs could be stored in the same directory structure at the location indicated by the save_path.

I could make a PR with this. What do you think?

[samet-akcay]

@LukasBommes, yeah this could be possible.

It's also possible to look at the results of the test set without running the inference script. So when you set log_images_to: ["local"] in project section of the config file, tools/train.py saves the results of the entire test set to `./results////images/...

Here is an example