TensorFlow-MIL

TensorFlow-MIL copied to clipboard

Deep Multiple Instance Learning

This repository contains a TensorFlow implementation of the CNN-MIL combination described in Classifying and Segmenting Microscopy Images with Deep Multiple Instance Learning from Brendan Frey`s lab.

Getting Started

• Install requirements (will install CPU version of TensorFlow): pip install requirements.txt.
• Follow instructions on (tf_cnnvis)[https://github.com/InFoCusp/tf_cnnvis] README. tf_cnnvis/ should be at the root after installation.

Options

The following options are available for running the model:

• -e, Number of epochs for which to train the model
• -r, Specify the seed
• -b, Batch size for training
• -s, Where to save model
• -m, Name of model
• -t, Whether to train (1) or load model (0)

Datasets

The datasets.py file contains a small cluttered MNIST dataset. Each image is 72 x 72 pixels and contains four numbers: three are 0's, the other one is a number 1 - 9 excluding 0. The locations of these numbers in the image are semi-random. This dataset is a much smaller version of what the authors use in the paper.