Predictions

This is a live demonstration of the Label-free Activation Map, an eXplainable Artificial Intelligence (XAI) method for Convolutional Neural Networks (CNNs). It generates a heatmap derived from the raw activation map extracted from the final convolutional layer. Unlike traditional methods, this approach doesn't depend on labels and necessitates only a single forward pass. In this demonstration, we employ a pre-trained ResNet50 model trained on the ImageNet dataset.

Debug

Available devices

Log