Faster-RCNN

Project for EE769 Machine Learning
Authors : Avineil Jain, Sravan Patchala

Code Structure

1. train.py creates, trains the model and stores it
2. test_frcnn.py tests the model "my_frcnn_try.hdf5" from test and saves it to test_results folder
3. keras_frcnn contains the various functions used

Note

The model has only been trained for 15% of its training time due to constraints of time and computation. However, since it produces decent results, we decided to stop training and add it to the repo. The main purpose of the project was NOT to beat some benchmarks, but to learn and understand object detection and implement a state-of-the-art technique for it! :)

VOCdevkit

The folder which should contain the data has not been added because of space constraints. However, the folder is only needed to train the model. To add it, download from -
http://host.robots.ox.ac.uk/pascal/VOC/voc2007/index.html