feat(metric): 🌟 add NIMA

[kargaranamir]

First I tried this code, but I couldn't find any pre-trained models for it. According to the code it uses one of the ResNet models. However, most of the codes on the Internet use the previous version of this repo, and the maintainers of this project assure that the previous version is still valid and working. The code that I use is based on MobileNetv2, and the pre-trained model URL can be found here.

It seems the deployed model on the Heroku app is different from the pre-trained one (issue8, issue28)

[kargaranamir]

DENSENET architecture gets a better result according to this repo. Also, we tested NIMA's different architectures on the website aesthetic dataset provided by labinthewild and came to the same conclusion. The current model downloaded from here.

[mplaine]

Minor tasks:

• magic numbers -> constants in m18_nima.py
• assign m18 to a category in metrics.json
• add a score section for m18_1
• run evaluator and update histogram images and score details for m18 in metrics.json