These experiments were run using MATLAB 2020a.
Toolboxes utilized:
- Deep Learning Toolbox
- Computer Vision Toolbox
- Parallel Computing Toolbox
Each matlab scripts, will train a segmentation model on the dataset corresponding to the filename. Within each file we define the network architechture as well as specify the hyperparameters.
The models directory contains the models that we obtained by executing the above scripts. To evaluate these models simply load them into matlab and execute any of the evaluate_network scripts.
Once the network is trained you can evaluate it using the script evaluate_network.m
Once the predictions have been made you can visualize the segmentation predictions and obtain a plot such as the following:
The above plot was produced by the network file net_75iou.mat.
The above plot was produced by the network file net_mnist.mat.