TL/DR

Using this data guess the number of objects in each headshot in this set.. Winners (most accurate to true values) will receive free headshots in GD. Closes 01.06.2025 12PM Pacific. Submissions via DM. ONLY PROTOGENS WILL BE DRAWN. Make sure the numbers are in the same order as on the test set.

LEAST SQUARES CONTEST 01

In machine learning a common way of training a model is for it to estimate the value of a function given an input. For instance, given a house and its characteristics, try to estimate a possible selling price on the market.

This is accomplished in a variety of ways, such as Linear Regression and Neural Networks, but usually they involve minimizing the error of the actual output with respect to the predicted output using a loss function.

The loss function is usually known as the Least Squares minimization, which means to minimize the total squared error. Most simple linear regression works this way.

Total squared error is the sum of the squares of the differences between components/coordinates of the target vector and the predicted vector. For instance, if for a given input the target is (4, 2, 0, 6, 9) and the predicted output is (6, 4, 1, 8, 7) the total squared error would be: (4 - 6)^2 + (2 - 4)^2 +(0 - 1)^2 + (6 - 8)^2 + (9 - 7)^2 = 17.

THE RULES

Over the past months I have created 12 Geometry Dash headshots. The first four have the number of objects disclosed: SET 1-4

The other 8 do not have the number of objects disclosed. Your task is to guess the number of objects in FOUR of those headshots (The headshots are pre-determined). The winner is the person with the smallest total squared error across the 4-dimensional vector. Ties are broken randomly. YOU MUST PROVIDE FOUR NUMBERS - ONE FOR EACH HEADSHOT

Please make sure to submit the vector in the order that the images appear in the Reddit posting.

You do not have to guess integer values. This contest is posted across multiple platforms. The four most accurate entries will be chosen. ONLY PROTOGENS WILL BE DRAWN - RUNNER UPS MIGHT BE CHOSEN.

Submissions should be via dm on the platforms the contest is posted on.

IMPORTANT

Objects in Geometry Dash are usually stuff like simple shapes like triangles and rectangles. More complex objects such as circles, hearts, stars, etc. exist, but nothing too out of the ordinary. Training data uses both simple and complex shapes. The warp action performs linear and affine transformations (sliding, rotating, scaling, shear) on the objects.

All objects used in the drawings are each of a single color only. Some GD objects might have shading, but not in these drawings. The only exceptions are gradient objects that vary in opacity but not brightness or saturation.

This is a collection of images containing most sprites in Geometry Dash 2.1 (containing objects but also player icons, UI, particles, etc.). Some symmetrical or repeating objects are only partially represented here. 2.2 adds a lot more objects but most are not used.

Please DM or ping if you would like any clarifications or help (don't ask me for information directly related to the object counts).

(This is my first time doing a contest like this, so be patient!)