Deep learning models determine their decision criteria automatically during training, making it difficult to explain the basis for their decisions. This opacity has led to AI models often being referred to as “black boxes.”
However, there is increasing demand for methods to provide explanations for the decision-making processes of AI models. Grad-CAM is one such approach that offers insights into model behavior.
In this report, we used Grad-CAM to generate heatmaps for 3D shape classification tasks performed with PointNet. These heatmaps visualize the regions of the 3D shapes that the AI model focuses on when making classification decisions. By plotting the key regions on 3D shapes, we aim to provide interpretability to the model’s classification rationale.
You can try out this Grad-CAM model on our demo site.
Astraea-Software Demo-Site
