This was the setup I used for my Honors Thesis at the University of Massachusetts, An Analysis of F-RCNN vs YOLO in Table Detection.
The keras-version of Faster R-CNN was originally pieced together by RockyXu66.
Tensorflow v.1.15 Keras v.2.3.1 h5py v.2.10.0 Python 3.7.0
Google Colab comes bundled with Jupyter Notebook Support & most Python distributions by default. This project has been setup to install the appropiate libraries, so no changes are needed.
To run this locally, I would recommend using an Anaconda environment. Officially supported downloads/distributions can be found at: https://www.anaconda.com/ The following distributions are required in your conda environment: Python, Tensorflow, Keras, h5py.
The data folder contains all the images that will be used during training and validation, as well as their annotated bounding boxes in an (x1, y1, x2, y2) format. You can execute the following files in the given order:
- 0_preprocess_data.ipynb
- 1_train_model.ipynb
- 2_test_model.ipynb
