Object Detection from Scratch
6 articles in this series
Object Detection from Scratch: Part 1 - Why This Project Is Worth Building
Why a Magic: The Gathering card detector is a serious engineering project, not a toy demo. Part 1 frames the product problem, the architecture, and the technical journey ahead.
AArthur Costa
Object Detection from Scratch: Part 2 - Dataset, Labels, and the Reality of Training Data
Part 2 examines the dataset behind the MTG detector: splits, YOLO labels, class design, annotation noise, and why data quality sets the real performance ceiling.
AArthur Costa
Object Detection from Scratch: Part 3 - Training the Detector
Part 3 explains how the MTG detector is trained: transfer learning, augmentation, optimization, early stopping, local CPU training, cloud GPU experiments, and the research behind each choice.
AArthur Costa
Object Detection from Scratch: Part 4 - Reading Metrics Like an Engineer
Part 4 explains how to read detection metrics in this project: precision, recall, mAP50, mAP50-95, per-class behavior, confusion patterns, and what the numbers imply for real product use.
AArthur Costa
Object Detection from Scratch: Part 5 - From Detection to Identification
Part 5 follows the live identification pipeline: detection, OCR, Scryfall lookup, DINOv2 art matching, error propagation, and why the system needs more than a detector.
AArthur Costa
Object Detection from Scratch: Part 6 - Shipping the System
Part 6 closes the series by looking at the web application, deployment boundaries, correction workflow, limitations, operational tradeoffs, and what it takes to ship the detector as a usable system.
AArthur Costa