SpeedOfSpin/Neural-IMage-Assessment

NIMA: Neural IMage Assessment

This is a PyTorch implementation of the paper Neural IMage Assessment by Hossein Talebi and Peyman Milanfar. You can learn more from this post at Google Research Blog.

The model was trained on the AVA (Aesthetic Visual Analysis) dataset, which contains roughly 255,500 images. You can get it from here. I used 80% of the dataset for training, and 5,000 images for validation. Note: there may be some corrupted images in the dataset, remove them first before you start training.
I used a VGG16 pretrained on ImageNet as the base network of the model, for which I got a ~0.22 EMD loss on the 5,000 validation images. Haven't tried the other two options (MobileNet and Inception-v2) in the paper yet. # TODO
The learning rate setting differs from the original paper. I can't seem to get the model to converge with momentum SGD using an lr of 3e-7 for the conv base and 3e-6 for the dense block. Other settings are all directly mirrored from the paper.
The code is written in Python 2, but should be portable to Python 3 without sweat.

Set --train=True and run python main.py to start training. The average training time for one epoch with --batch_size=128 is roughly 1 hour on a Titan Xp GPU. For evaluation, set --test=True instead.
I found https://learning-rates.com/ a very handy tool to monitor training in PyTorch in real time. You can check it out on how to use it. Remember do pip install lrs first if you are inclined to use it.

Here shows the predicted mean scores of some images from the AVA dataset. The ground truth is in the parenthesis.

The predicted aesthetic ratings from training on the AVA dataset are sensitive to contrast adjustments. See below (bottom right is the original input).