Empirically-based Features

Here are the codes for our empirically-motivated features that designed for encoding human perception of realism.

Pre-trained CNN Models

We share the first test of pre-trained CNN (VGG-19) using Keras with a Tensorflow backend. 
Model for realism classification.
Model for realism prediction (regression).

Pre-trained Multi-layer Perceptron (MLP) Models

We share the first test of pre-trained MLP using Keras with a Tensorflow backend.
Model for realism classification.
Model for realism prediction (regression).

Codes for CNN Models and MLP Models

We share the codes of jupyter notebook which used to train and test. To successfully run the codes, you may need to install Tensorflow 1.2 and Keras 2.0.5.

Reference

The codes and CNN models are developed for following papers. Please cite our paper if you find the codes and models useful in your research.

S. Fan, T. -T. Ng, B. Koenig, J. Herberg, M. Jiang, Z. Shen, Q. Zhao. “Image Visual Realism: From Human Perception to Machine Computation”. IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), accepted, 2017.

S. Fan, T.-T. Ng, J. Herbert, B. Koenig, C. Tan, R. Wang, “An Automated Estimator of Image Visual Realism Based on Human Cognition”, IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2014.

Please email Zhiqi Shen at dcsshenz@nus.edu.sg if you have any questions or comments.