← All projects

Human Digit Classification

Jan – Mar 2025
PythonJavaScriptHTMLPyTorchML

Highlights

  • Built a two-stage CNN pipeline in PyTorch: bounding-box regression with Generalized IoU loss → digit-count classification; improved F1-score through iterative evaluation.
  • Created a custom, fully annotated dataset of 10,000+ images from scratch using home-grown labeling tools.
  • Applied image + annotation transformations for data augmentation and robustness.