classification

Repository for emDNN compressed classification models.

CIFAR-10 classification

• Classification using emDNN compressed ResNet-18 can be done using:

  python cifar_classifier.py -p models/resnet18/resnet18_emdnn.prototxt -m models/resnet18/resnet18_emdnn.caffemodel
  

• To evaluate pretrained model use normalized CIFAR-10 images, using mean = (0.4914, 0.4822, 0.4465) and scaled down by 255.0 and save it to a npz file to be read by eval_cifar_caffe_model.py or else download already normalized CIFAR-10 from here.