Fully Convolutional Networks (FCNs) for Image Segmentation

A post showing how to perform Image Segmentation using Fully Convolutional Networks that were trained on PASCAL VOC using our framework. In this post we want to present Our Image Segmentation library that is based on Tensorflow and TF-Slim library, share some insights and thoughts and demonstrate one application of Image Segmentation. To be more precise, we trained FCN-32s, FCN-16s and FCN-8s models that were described in the paper "Fully Convolutional Networks for Semantic Segmentation" by Long et al. on PASCAL VOC Image Segmentation dataset and got similar accuracies compared to results that are demonstrated in the paper. We provide all the training scripts and scripts to convert PASCAL VOC into easier-to-use .tfrecords Moreover, it is very easy to apply the same scripts to a custom dataset of your own.