Black.ai Computer Vision Technical Challenge

The purpose of this challenge is to test your ability and chosen approach in dealing with a computer vision problem.

Introduction

You are tasked with developing an algorithm to refine object detections from partially segmented image data.

The task

1. Download the sample input data from here or using the cli:

   wget -t 15 https://s3-ap-southeast-2.amazonaws.com/public.data.black.ai/cv_challenge_dataset.zip

   The dataset contains a series of the following:

   1. A depth image with a scene containing people.

   2. A label mapping for this image. The Mapping represents regions containing people as:

      • 0 = furniture
      • 1 = floor
      • 2 = walls/ceiling
      • 3 = female
      • 4 = male

2. Using any language or collection of libraries you wish, write a short program that for each depth image:

   • Loads the input data
   • Applies the relevant label mask to the depth image;
   • segments the masked depth image;
   • Handles occlusion in the label mask; and
   • Outputs the number of unique people in each of the given input images.
   • Outputs the time taken to process this image in miliseconds.

   You can go about this in any way you wish, though your implementation should be robust to the following scenarios:

Scenario 1: Separate clearly defined people

Scenario 2: Occlusion; Overlapping people

Scenario 3: Occlusion; Incomplete label mask

Scenario 4: Erranous labelling artifacts.

ie. noise whereby labels encapsulate impossibly small regions.

Submission

Just upload your codebase to a public github repository and send us the link!

Final submission should be made to [email protected], with the subject line "CV_CHALLENGE | <your_name>"

note: please also include instructions for simple setup and testing, so that we can easily run your program against our own test data. Be sure to include a list of dependencies, if any are required, to simplify the installation process.

Evaluation

We are primarily looking to assess the accuracy of detection, detection speed, and the implementation itself. We will be testing/training both the implementation and model on a machine running 64-bit Ubuntu 16.04 LTS, with a quad-core CPU, and a Nvidia GTX 1080 graphics card.

Bonus

Alongside the number of unique individuals in any given input image, report any additional information you can think of for each person.

Happy hacking!