Challenge has Ended

COVID-19: Social Distancing AI Challenge

Detect people not following social distancing norms.


Bounty for Rank 1
67 of 100 Participants

The deadly COVID-19 is spreading exponentially and is causing havoc in our lives. Social distancing and wearing face masks are the only effective strategy against it. You as data scientists have a chance to play a key role in this battle. The task is to detect people in the images in the dataset and classify them into two classes:

  1. Safe Distance
  2. Distance Violation

The provided dataset is Mall Dataset, crowd counting dataset. The mall dataset was collected from a publicly accessible webcam for crowd counting and profiling research. Ground truth: Over 60,000 pedestrians were labelled in 2000 video frames. The data was annotated exhaustively by labeling the head position of every pedestrian in all frames. The images are provided ".jpg" format. The annotations are provided in "mall_gt.mat", there is an additional file "Visualizing (mat) files.ipynb" to help visualize the annotations on the images.

The dataset contains head positions of the pedestrians. Based on their head position the person has to be classified using the formula:

threshold_width = (y co-ordinate of head position)/3

threshold_height = (y co-ordinate of head position)/6

 if there is another head position within the horizontal ellipse centered at the head position, having parameters width 'threshold_width' and height 'threshold_height' classify as : "Distance Violation"

else: "Safe Distance"

Participants will have to account for corner edge cases for the ellipse.


This hackathon is conducted currently for research purposes, and no commercial use will be made for the same.


  1. From Semi-Supervised to Transfer Counting of Crowds
  2. Cumulative Attribute Space for Age and Crowd Density Estimation
  3. Crowd Counting and Profiling: Methodology and Evaluation
  4. Feature Mining for Localised Crowd Counting

The dataset contains "Frames" directory containing ".jpg" images sequences, Ground Truth "mall_gt.mat" and other relevant files. More information regarding the dataset is provided in the "readme.txt". The links for the dataset and relevant research is given in the "About" section of this hackathon page.

The output of the model should be head positions for person co-ordinates in an image and their associated class ("Safe Distance", "Distance Violation").

The final submission should include:

  1. Model Files (in formats supported by the frameworks given below)
  2. Training Files (.ipynb, .py)
  3. requirements.txt (providing details of modules required to run your submission)
  4. Code Execution file (

Other files may be included for purposes of code modularity.

The file should be named “” and will be supplied a “--testdir” named command line argument for evaluation, which will contain the images in "jpg" format, the example is given below: —testdir <path to test directory>

The output file should be named "output.json" having the nested format given below:

{"<file name 1>":

       {"location of head position1 (x, y)" : "<class>",

     "location of head position2 (x, y)" : "<class>", 

          "location of head position3 (x, y)" : "<class>", 


"<file name 2>":

       {"location of head position1 (x, y)" : "<class>",

     "location of head position2 (x, y)" : "<class>",   






  “1.jpg”: {

    “(10, 200)“: “Safe Distance”,

    “(200, 300)“: “Safe Distance”


  “2.jpg”: {

    “(10, 200)“: “Safe Distance”,

    “(200, 300)“: “Safe Distance”  }}

Supported Frameworks:

Currently only Python-framework based models are accepted:

  • Tensorflow
  • Pytorch
  • Keras
  • Caffe2
  • Onnx

The Judgement

The criteria of judgement will be:

  1. Accuracy of detection algorithm
  2. Accuracy of safe distance classification
  3. Speed - Accuracy trade off


  1. Submission must not include copyrighted code. If violation is found, submission will be rejected.
  2. The submission should be in a proper format as described by README.txt (included in the dataset).
  3. Late submission will not be accepted beyond provided deadline (Indian Standard Time).


The winner of the hackathon will be awarded INR 15,000.

Jury Award - worth INR 5000

One of the participants will be awarded "Jury Award" for applying innovative methods for problem solving.

The Jury Award winner will receive goodies including:

  • T-shirt
  • Premium Laptop Bag
  • Personalized Mugs
  • Personalized Notepads
  • Laptop Stickers

All the participants with "Successful Submission" will receive certificates from "Dockship".

How do I apply for this Challenge?
How do I download the dataset?
Can I make multiple submissions?
Where will the results be declared?
Can we apply as a team?
I've other queries, where can I get support?
Challenge Announced
07-Jul-2020, 11:20 am IST
16-Jul-2020, 6:00 pm IST
Challenge Started
Application Deadline
29-Aug-2020, 6:00 pm IST
31-Aug-2020, 6:00 pm IST
Challenge Ended