Object Detection with Bounding Box Regression

The project demonstrates the working of the Bounding Box Regression technique used in object detection tasks. It can efficiently predict the four coordinates of a bounding box around the image with its class probabilities too.

View the article on Medium ~> Getting Started With Bounding Box Regression In TensorFlow

Check out the Google Colab notebook ~> https://colab.research.google.com/drive/1usT_XYE6DLENeUL3__GNCYAWn-0NbVh6#forceEdit=true&offline=true&sandboxMode=true

The following files are included in this repo:

1. DataProcessor.py : Extract the images and XML annotations to convert them .npy files ready for training/testing.

2. Model.py : Defines the CNN model and other useful methods.

3. MainFile.py : Trains the model on the data.

4. Evaluation.py : Loads a model from the given h5py file and predicts bounding boxes for various images, draws them on the image and then finally saves the images to a directory.

By default, the Evaluation.py file reads the pretrained model weights which are included with the repo.

Make sure you download the data first -> https://www.kaggle.com/mbkinaci/image-localization-dataset