Capsule Network for Road Sign Classification

Description

This project is focused on the problem of traffic sign recognition, where it adapts and tunes the Capsule Network architecture for this particular classification task. The resulting architecture is evaluated on different benchmarks and explained with the LIME approach of Explainable AI.

Instructions

Reproduction:

1. Clone the repository. Let us call to the root folder project_root, i.e. all python files are located in project_root/capsnet.
2. Download benchmarks from
   https://drive.google.com/file/d/10Nl3ucj1b4u8H-yvl3JNTIBWLtiVN2Eu/view?usp=sharing
3. Unpack the contents in the folder project_root. The archived folder benchmarks should come right into the directory project_root. I.e. the folder tree should look project_root/benchmarks/{belgium-TSC,china-TSRD,german-GTRSD,rtsd-r1}.
4. Install prerequisites. For Anaconda under Windows, an option is to run the script install.bat inside project_root/capsnet.
5. Navigate to project_root/capsnet. Call python run_capsnet.py --benchmark all (alternatively, launch.bat) and wait until execution is over. Results will be stored in the folder project_root/capsnet/traindir.
6. One can open traindir with TensorBoard by calling tensorboard --logdir=./traindir from project_root/capsnet. Execution summary is available at project_root/capsnet/traindir/stats.json.

Already processed traindir with execution results is available at: https://drive.google.com/file/d/1enFqS_TE7eQ5wB3AKLoJ3sTtjVFiuKgP

Other functionality:

1. Script executable.bat prepares standalone executable.
2. Script document.bat prepares Sphinx-processed documentation.