# Image Classification — Test Project Configuration
#
# Data size: 50 images (64x64 RGB PNG) in 3 subdirectories (cats/dogs/birds)
# Download: python dev/test-datasets/download.py --dataset images
#
# Project name:
images-test

# Data path:
/home/mrichardson/Projects/Urika/dev/test-datasets/images/data

# Description:
Tiny synthetic image classification dataset for testing Urika's source scanner, data
profiler, and image-handling capabilities. The dataset contains 50 small (64x64 pixel) RGB
PNG images organized into three subdirectories representing classes: cats (17 images), dogs
(17 images), and birds (16 images). Each image consists of random pixel patterns with a
class-specific color tint (warm orange for cats, blue-purple for dogs, green for birds) and
overlaid geometric shapes for texture variety. These are not real photographs — the color
tinting provides a trivially learnable signal, making the classification task achievable
even with simple pixel-based features. The purpose is to verify that Urika can detect image
file types, profile directory-structured datasets, and orchestrate a basic image
classification pipeline. Success is achieving high classification accuracy (>90%) given the
clear color separation between classes.

# Research question:
Can a simple image classification pipeline distinguish between the three classes (cats,
dogs, birds) based on color and texture features, and what classification accuracy is
achievable?

# Mode:
exploratory

# Web search:
no

# Venv:
yes

# Knowledge suggestions:
Add the data-description.md from dev/test-datasets/images/knowledge/. No additional
knowledge needed for this synthetic test dataset.
