Object Localization

This project was done as a part of a Hackathon hosted by Flipkart, by the name Flipkart GRID Challenge - Teach the Machines (powered by Dare2Complete).

As the name suggests, the goal was to put a bounding box around an object in the image.

We were provided with:

1. 14000 640x480x3 coloured training images (each image had one object)
2. Image Labels corresponding to each image. (labels were the x,y coordinates of the top-left and bottom-right corners of the box)

We were supposed to maximize the IOU (intersection over union) of our predicted bounding box. We were able to push that figure (IOU) upto 0.83.

Competition was from Feb, 2019 to March, 2019. My friend Divyank Pandey was in it with me.

So I thought of uploading our work and maybe making it a little more reproducible.

Results

  Round - 3:                                 Round - 2:

You can download:

1. Prepared Dataset
2. Pretrained models:
   • Round - 3
   • Round - 2

Refer Approach Text(s) for Round - 2 and Round - 3 for understanding the correct way of using the above.