About

This interactive demo showcases the Label-free Activation Map for Convolutional Neural Networks. It uses a ResNet50 model trained on the ImageNet dataset (1000 categories) and generates heatmaps from the model's final convolutional layer without relying on labels.

Interactions

Use your camera, upload an image, or select a pre-defined image from the dropdown menu.

• Pause Camera: Essential for interacting with the grid.
• Grid Click: Click on specific image regions to see what the model focused on.
• Prediction Click: Click a class name to highlight where that object was detected. Clear selection by clicking again or clicking the X-button.
• Mode Selection: Choose different model modes to switch between label-free/Grad-CAM heatmaps, multi-instance classification, and object detection.