Interpretable Deep Vision Model Enhancing Robustness and Transparency in Robotic Perception

This dataset contains the performance evaluation metrics of an interpretable deep vision framework designed for robust and transparent robotic perception. Three models were evaluated: a baseline CNN, a CNN with data augmentation, and the proposed interpretable deep vision model. The metrics include classification Accuracy, Precision, Recall, p-values for statistical significance, IoU (Intersection over Union) Scores, and 95% confidence intervals for localization consistency. Experiments were conducted under dynamic environmental conditions including varying illumination, occlusion, and background complexity. The dataset supports reproducibility and analysis of the comparative performance of interpretable AI models for robotics applications.