Hardware Vision Challenge #1: Tabletop Object Counting & Segmentation

Name: Hardware Vision Challenge : Tabletop Object Counting
Start: 2026-09-01T07:24:46+00:00
End: 2026-10-04T23:59:59+00:00

Welcome to the Hardware Vision Challenge! In this competition, your goal is to build machine learning and computer vision models to detect, segment, and count physical tabletop objects (Books and Glasses) under variable studio lighting, shadows, orientations, and background textures.

Objective

You are provided with a dataset of 2,000 synthetic tabletop scenes (512 x 512 RGB resolution). For the 400 test images (1601.png to 2000.png), your task is to predict the count of: - books: an integer count between 0 and 5 - glasses: an integer count between 0 and 5

Evaluation Metric

Submissions are evaluated using Mean Absolute Error (MAE) across all target object count predictions (books and glasses). The formula for MAE is:

MAE = (1 / 2N) * sum from i=1 to N of ( |y_hat_i,books - y_i,books| + |y_hat_i,glasses - y_i,glasses| )

where N = 400 test images. Lower MAE scores indicate higher precision.

Submission Format

Your submission file must be a CSV file containing the id of each test image and your predicted counts for books and glasses. The file should contain a header and have the following format:

id,books,glasses
1601,2,1