One hundred very busy scientists are working around the clock to develop a machine that will cure all major illnesses, deep down in a secret lab. They often walk up to the machine, pick up one of its components, or put another one back. Occasionally, they wave to their colleagues to signify that something isn't working well.
You are the foreman of this highly secretive laboratory and your job is to note all the actions that the scientists are performing in real time. However, there is only one of you and one hundred of them, so you decide to create a machine that can do this work for you.
Using either the most creative or effective technique you can think of, develop a system (doesn't have to rely fully, or at all, on DL) to classify the scientists' activities.
Creative solutions, whether they are failures or not, will still be invited for interviews.
Download the training set here
Download the test set here
Each csv file is a collection of keypoints for a skeleton performing a single action over a series of frames.
There are 5 different actions, one passive (standing) state and:
Each csv file contains `frame_number, keypoint_number, x, y, confidence_score`.
The keypoint_number is a number between 1 and 18, using MSCOCO's keypoint definition.
A keypoint can be missing for a specific frame if occluded, `x` and `y` is the coordinate of a specific keypoint.
The confidence_score is a number between 0 and 1, indicating how sure we can be that the current keypoint is present.
There are no limitations on the way you're allowed to solve the challenge. When it comes to validating your model, you should prepare a result csv file, mapping each file in `dataset/test/` to an action. Use the following format:
Send an email to firstname.lastname@example.org with your results attached. Make sure you set the subject to recruitment challenge. If you have additional questions or require an extension, email us.