cmuchancel/hw1-pool-balls

24-679 (Fall 2026): Pool Balls — Solid vs. Stripe

cmuchancel/hw1-pool-balls

Student-created photographs of pool balls on a blue background, prepared as square RGB images with repeated training variants. The image-classification task predicts whether a pool ball is solid or striped.

Source and task

The preparation notebook uses a ZIP containing solid and stripe image folders. The class label comes from the folder containing each photograph.

0 = solid / no visible white stripe; 1 = stripe / visible white stripe.

The source dataset contains 30 original photographs: 16 solid images and 14 stripe images. The student photographed pool balls on a blue background and organized the images by class. Ball numbers are not targets, and no fixed two-photograph pairing is assumed.

Course: 24-679, Fall 2026, Carnegie Mellon University. Repository maintainer: the account shown above.

Fields

Field	Meaning and modeling role
image	Prepared RGB pixels; the model's image input.
label	Classification target: 0 = solid; 1 = stripe.
label_name	Readable target description; exclude from model inputs.
source_id, parent_id	Unique prepared/augmented example key and its original source key; provenance only.
augmentation, is_augmented	Method identifier and random-augmentation flag; provenance only.

Splits and original-source counts

These counts are computed from the packaged splits for this run.

Split	Original rows	Synthetic rows	Total rows
train	21	336	357
validation	4	0	4
test	5	0	5

Class counts include synthetic rows in training only.

Split	Label	Rows
train	solid	187
train	stripe	170
validation	solid	2
validation	stripe	2
test	solid	3
test	stripe	2

Requested holdout fraction: 30%; test receives 50% of that holdout. Small-sample rounding can change the realized image-level proportions. The first split uses seed 24679, and the holdout split uses seed 24680.

Following the instructor's September 15 split clarification, original photographs are split with class stratification before augmentation. Training includes its originals and their variants; validation and test contain only unaugmented prepared originals. A source and its descendants never cross boundaries. Different photographs of the same physical ball may appear in different splits, so these partitions do not measure generalization to unseen physical balls.

Augmentation and preprocessing

Prepared image size in this run: 224 × 224 RGB. Padding color is (128, 128, 128).

All splits receive the same deterministic preparation: apply EXIF orientation, convert to RGB, resize while preserving aspect ratio, and pad to the configured square size. Working copies discard camera metadata; native source files remain separate.

Each original training image contributes its prepared parent and 4 independent variants from each of four methods. Transforms are not stacked.

The notebook uses brightness factors 0.4–0.7 or 1.4–2.0; rotation of 15–30 degrees in either direction on a fixed canvas; contrast factors 0.35–0.65 or 1.5–2.2; and Gaussian blur radius 2.0–4.5 pixels.

The stored identifiers mild_brightness, small_rotation, mild_contrast, and mild_gaussian_blur are retained from the classroom notebook.

Training method	Stored rows
mild_brightness	84
mild_contrast	84
mild_gaussian_blur	84
none	21
small_rotation	84

Intended use and limitations

Use for teaching image preparation, augmentation inspection, provenance, and binary image classification. The collection is small and visually controlled. The blue pool-table background, lighting, camera position, and repeated physical balls may act as shortcuts.

Synthetic variants do not add independent physical objects or independent photographs. Strong brightness or contrast changes can hide details, rotation may crop evidence, and blur can make the stripe less visible. Copied labels and provenance assertions do not prove visual label preservation, so the notebook includes a parent/variant gallery with both classes for inspection.

The dataset should not be interpreted as measuring performance on arbitrary billiards photographs with different tables, lighting conditions, camera angles, or ball sets.

Privacy and licensing

The photographs are intended to contain pool balls and no people or faces. The packaged splits omit the original upload filenames, and the prepared images do not retain camera metadata.

The original ZIP remains a separate source artifact. Review applicable course permissions before reuse or redistribution. License: all rights reserved; no reuse license is granted.

AI usage disclosure

ChatGPT was used to help adapt the provided 24-679 lecture notebook to this pool-ball dataset and to assist with code and documentation. The 30 original photographs were created by the student and were not generated by AI. Synthetic images are produced only by the explicit programmatic transformations shown in this notebook.

Load and compare

from datasets import load_dataset
ds = load_dataset("cmuchancel/hw1-pool-balls")
# Train with ds["train"], choose settings with ds["validation"], then score ds["test"].

Use an account with access if repository visibility changes. For reproducible comparisons, record the dataset commit and model/environment versions. Regenerate this card with the preparation notebook after changing the data; its counts are calculated from the actual packaged splits.