Famous Paintings

Famous Paintings

Context

When we refer to famous paintings, we must also consider objective characteristics such as canvas size, the type of brush used for painting, visual complexity, or color density, alongside their artistic quality. Automatic value estimation and painter style identification can help collectors and museums analyze large collections of artworks.

To solve this problem, you will have two files at your disposal:

• train.csv – contains all variables, including target_price, for training models
• test.csv – contains the same variables, without the target_price column, and the data in this file is used for generating predictions

Dataset Description

Requirements

For this problem you have to solve 3 tasks.

Task 1 – Artistic Authenticity Score (20 points)

Recently, a revolutionary algorithm has been discovered by a group of researchers in artificial intelligence and art. It can evaluate, based on the objective characteristics of the painting, whether a painting is certainly authentic or not.

For each painting, the algorithm calculates an Artistic Authenticity Score (AAS) using the following rules:

1. If stroke_density > 0.7, add 2 points to AAS.
2. If complexity > 0.65, add 2 points to AAS.
3. If uses_gold_leaf = True, add 1 point to AAS.
4. If has_signature = True, add 1 point to AAS.
5. If num_colors > 65 and colorfulness > 0.7, add 2 points to AAS.
6. If contrast < 0.4 or brightness < 0.45 or brightness > 0.75, subtract 1 point from AAS.

After calculating the score, the painting receives the label:

• Autentic: if AAS ≥ 5

• Incert: if AAS < 5

• Answer in the submission file will be Autentic or Incert for each painting in test.csv.

Task 2 – Grouping paintings by painter style (40 points)

It is known that the paintings in the dataset were created by 5 famous painters, but the information about the painter is not available.

Your task is to assign each painting in test.csv a numeric ID between 0 and 4, so that paintings created by the same painter receive the same ID.

• Answer in the submission file will be the numeric ID (0, 1, 2, 3, or 4) for each painting in test.csv.
• All paintings created by the same painter should have the same ID.

Task 3 – Painting price estimation (40 points)

For each painting included in train.csv we have a specified price (the target_price column). Based on the information available in train.csv, train an Artificial Intelligence model that can predict the price for the paintings in test.csv.

Submission File Format

The submission.csv file must contain:

• SampleID - taken from the test.csv file;
• subtaskID - can be: Task1, Task2, Task3;
• Answer - the calculated or estimated value for the painting indicated by SampleID for the task indicated by subtaskID.

Evaluation

Subtask 1 – Authenticity score

For this subtask, exact comparison is used between submitted answers (Autentic / Incert) and correct labels.

• Maximum score is awarded only if all answers are correct.
• If there are incorrect answers, the score is awarded proportionally, using the rule of 3 simple.

Subtask 2 – Painter grouping

Evaluation is done using Adjusted Rand Index (ARI).

• Maximum score is awarded if:

• For values below this threshold, the score is awarded proportionally.

Subtask 3 – Price prediction

Evaluation is done using Mean Absolute Error (MAE).

Rules:

• MAE ≤ 600 → maximum score
• MAE ≥ 2000 → 0 points
• For values between these two thresholds, the score is awarded proportionally.