Miles Deep is a video classification tool that uses a deep convolutional neural network with residual connections, built on the Caffe framework, to classify each second of video into 6 categories with 95% accuracy. It then automates video editing based on those classifications, such as removing scenes without sexual contact. The system supports GPU acceleration via CUDA and CuDNN for faster processing.