Skip to content
An object localizer made for the Flipkart grid ML challenge
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.


An object localizer made for the Flipkart grid ML challenge. We achieved 86% accuracy(approx), and had a rank of 124 in the contest.

Getting Started

These notbooks were made and imported from Google Colab. The following notebooks would run as it is in colab. To run the same in your own machine, download the datasets into a directory and provide and change the path variable to that directory and you are good to go.


Link to the Google drive link for the train dataset :- Link to the Google drive link for the test dataset :-

Use the image names in the CSV files in the repository to access the images.


alt text


NOTE: If running natively on your machine set the path to the directory where the images are stored.

Run the cells in the DatMan.ipynb file. This file takes 3500 images from the training dataset and stores it as a .np file. Use this file in the FChal2.ipynb file, to train the model and save the model. Using this saved model, generate a .csv file using the TestGenerator.ipynb.

In order to check your predictions on the images itself, run the designated cells in the FChal2.ipynb file.

The code, in particular was tested in the steps given above. To run a train and test generator based model (would run faster natively in your machines but slower on Google Colab).


In the given problem, the Resnet-5 model has been used where the input is an array of 244x244x3 numpy array and we map it to 4 output cells, each denoting the coordinates (x, y, width, height) of the bounding box.


The model is inspired by:-


Hersh Dhillon, Harshit Malik, Shreekanth Ajith

You can’t perform that action at this time.