This is my first and possibly most successful object detection(without classification) script until now. This is created using Tensorflow, OpenCV, and some other helper libraries. Everything including the preprocessing, model creation, post processing and evaluation is done from scratch. The code is super easy to understand. The annotated dataset can be found on kaggle.
The link to the code is as follows: https://colab.research.google.com/drive/1C5Nx7Vu80HKWvRE8b-zuw0hcZsrWdKMR?usp=sharing
The images and bounding boxes are plotted with and without Non Max Suppression(NMS).
Could not complete the epochs specified as GPU ran out but will surely try to train the full model