This project evaluates the performance of the MiDaS model for monocular depth estimation on the DIODE Outdoor dataset. It includes:
• Loading and preprocessing DIODE Outdoor data
• Running MiDaS on selected test images
• Computing evaluation metrics such as RMSE, MAE, AbsRel, and δ accuracy
• Visualizing prediction errors and depth maps
• Analyzing model failure cases (e.g., sky, glare, shadows)
The notebook is designed for Google Colab and uses PyTorch with timm and OpenCV libraries.