v1.1.0 — Tests, Documentation & Visualizations

Added

• pytest test suite covering model output shapes, gradient flow, NaN-aware loss edge cases, dataset utilities, and config loading
• Sample prediction images, training curve visualizations, and model comparison overlay in assets/

Changed

• Rewrote README with badges, embedded images, architecture diagrams, and code examples
• Added parameter counts and strategy details to results table

Fixed

• Deprecated df.fillna(method="ffill") replaced with df.ffill() for pandas 2.x compatibility

v1.0.0 — Clean Modular Rebuild

Complete modular rebuild merging both CNN and ResNet implementations into a single package with shared utilities, YAML config, NaN-aware loss, and dual-framework support (PyTorch + Keras).

v2.1.0

Loss Function Improvement:
Using a DIY MSE loss function that ignores nan
and only calculates loss for keypoints that are not nan
without dropping any sample data

v2.0.0

Model Improvement:
Using Residual Neural Network instead
Removed Data Augmentation as it seems to be not improving the results