Round 1: Completed #educational Weight: 30.0

# SKELY

Recognize Orientation of a 3D Skeleton

3010
383
34
392

🛠 Contribute: Found a typo? Or any other change in the description that you would like to see? Please consider sending us a pull request in the public repo of the challenge here.

## 🕵️ Introduction

Where is the skeleton facing ?

This problem is an adapted version of the ORIENTME which adds a few more layers to complexity to the task of predicting the orientation of a 3D model from a single picture.

This problem introduces a non-uniform distribution of samples in the training and the test set. And more importantly, this problem also evaluates out-of-distribution generalization.

For input you will be given a large number of images, for about half of them, we have measured the orientation of the skeleton. But to be able to stitch all those images together, you have to figure out how to predict the orientation of the skeleton for the rest of the images.

Understand with code! Here is some code to get you started right away ! 😄

## 💾 Dataset

The training dataset consists of 9999 images of size 2048x1898 with 4 channels each (for RGBA). The associated label is a single continuous variables :

• xRot : Orientation of the Skeleton, in degrees, along an arbitrarily chosen axis (a number between 0 and 360). The axis around which this value is measured, is consitent across the whole of the training and the test set.

The test dataset consists of 10001 images of size 2048x1898 with 4 channels each (for RGBA). The goal of the task is to predict the xRot value of the Skeleton in these test images.

## 📁 Files

Following files are available in the resources section:

• train.tar.gz - (9999 samples) Tar File containing all the training images, and associated labels

• test.tar.gz - (10001 samples) Tar file containing all the test images

• sample_submission.csv - A sample submission file (with random predictions) to demonstrate the expected file structure by the evaluation setup.

## 🚀 Submission

• Prepare a csv containing header as :
• filename, xRot
• The values of the filename column should match the file name of each of the images in the test set.
• The values of the xRot should be a number between 0 and 360, representing the orientation of the skeleton (in degrees) along the respective axis.
• Sample submission format is available in the resources section of the challenge page as sample_submission.csv.

Make your first submission here 🚀 !!

## 🖊 Evaluation Criteria

During evaluation Mean Squared Error be used to test the overall performance of your solution.