Dynamic Early-Exit Convolutional Neural Networks for Edge Vision: Do only what and when you need!

While computer vision has seen strong growth due to deep convolutional neural networks, deploying these models on edge devices remains challenging due to constraints in power, computation, memory, and performance. This talk presents dynamic deep CNNs that adapt computation based on input complexity. By selectively activating layers, channels, or early exits, these models reduce
unnecessary computation, improving efficiency
for resource-constrained edge applications.