gamma-net

We used a convolutional neural network to predict γ-peaks based on the input image. In total, training data consisted of 31988 image patches, by pooling across monkeys, channels and sessions. We used all image patches from a unique 10% of the stimuli as a test set and used the rest for training. We resized the 224x224 image patches to 84x84 in order to reduce the number parameters in the network. We used the VGG-16 model from keras applications, with frozen weights pre-trained on ImageNet for transfer learning. The VGG-16 activations were the input to a network that consisted of 2 convolutional layers and a readout layer. We compared predictions from different VGG-16 input layers, and for the main examples of RFs used conv3 3 as input (Figure 6). Convolutional layers consisted of (3x3) filters with bias, stride (2), valid padding, L1-norm kernel regularization (0.001), leaky ReLU activation (0.1), and dropout (0.5). The final two convolutional layers had 32 and 16 units. The readout layer consisted of (4x4) filters with bias and leaky ReLU activation (0.1).

For more details, please see the methods part of our paper:

Predictability in natural images determines V1 firing rates and synchronization: A deep neural network approach

Installation

Clone/Fork the repository to use the scripts.

git clone https://github.com/uranc/gamma-net.git

Requirements

tensorflow v1.14, keras v2.2.4, scikit-image==0.17.2 (optional: other versions should also work)

pip install tensorflow==1.14
pip install keras==2.2.4
pip install scikit-image==0.17.2

Usage

Command-Line

You can use the pre-trained model based on VGG-16 to predict gamma peak value in log10 space.

Input size is fixed to be 84x84x3, training set consisted of Black & White images. Input can either be:

• A single image file (84x84x3) - see skimage.io.imread for details
• A 4D numpy array ( num_images x 84x84x3)

python pred.py --input examples/sample_im.png

Output will be printed on the command line.

python pred.py --input examples/samples_nat_im.npy

Output will be saved to examples/samples_nat_im_pred.npy

Jupyter notebook

Please check our notebook on how to use or modify the usage of the model: samples_nat_im_demo.ipynb.

New Features

• directory of images
• upgrade to tensorflow v2