Proof_verify.py

# -*- coding:utf-8 -*-
# Experimental proof for our Eq.5
import argparse
import cv2
import numpy as np
import torch
from torchvision import models

class FeatureExtractor():
    """ Class for extracting activations and 
    registering gradients from targetted intermediate layers """

    def __init__(self, model, target_layers):
        self.model = model
        self.target_layers = target_layers
        self.gradients = []

    def save_gradient(self, grad):
        self.gradients.append(grad)

    def __call__(self, x):
        outputs = []
        self.gradients = []
        for name, module in self.model._modules.items():
            x = module(x)
            if name in self.target_layers:
                x.register_hook(self.save_gradient)
                outputs += [x]
        return outputs, x


class ModelOutputs():
    """ Class for making a forward pass, and getting:
    1. The network output.
    2. Activations from intermeddiate targetted layers.
    3. Gradients from intermeddiate targetted layers. """

    def __init__(self, model, target_layers):
        self.model = model
        self.feature_extractor = FeatureExtractor(self.model.features, target_layers)

    def get_gradients(self):
        return self.feature_extractor.gradients

    def __call__(self, x):
        target_activations, output = self.feature_extractor(x)
        output = output.view(output.size(0), -1)
        output = self.model.classifier(output)
        return target_activations, output


def preprocess_image(img):
    means = [0.485, 0.456, 0.406]
    stds = [0.229, 0.224, 0.225]

    preprocessed_img = img.copy()[:, :, ::-1]
    for i in range(3):
        preprocessed_img[:, :, i] = preprocessed_img[:, :, i] - means[i]
        preprocessed_img[:, :, i] = preprocessed_img[:, :, i] / stds[i]
    preprocessed_img = \
        np.ascontiguousarray(np.transpose(preprocessed_img, (2, 0, 1)))
    preprocessed_img = torch.from_numpy(preprocessed_img)
    preprocessed_img.unsqueeze_(0)
    input = preprocessed_img.requires_grad_(True)
    return input


class GradCam:
    def __init__(self, model, target_layer_names, use_cuda):
        self.model = model
        self.model.eval()
        self.cuda = use_cuda
        if self.cuda:
            self.model = model.cuda()

        self.extractor = ModelOutputs(self.model, target_layer_names)

    def forward(self, input):
        return self.model(input)

    def __call__(self, input, index=-1):
        if self.cuda:
            features, output = self.extractor(input.cuda())
        else:
            features, output = self.extractor(input)

        if index == -1:
            index = np.argmax(output.cpu().data.numpy())

        synset = [l.strip() for l in open('./synset.txt').readlines()]
        print("class of interest: ", synset[index])
        one_hot = np.zeros((1, output.size()[-1]), dtype=np.float32)
        one_hot[0][index] = 1
        one_hot = torch.from_numpy(one_hot).requires_grad_(True)
        if self.cuda:
            one_hot = torch.sum(one_hot.cuda() * output)
        else:
            one_hot = torch.sum(one_hot * output)

        self.model.features.zero_grad()
        self.model.classifier.zero_grad()
        one_hot.backward(retain_graph=True)

        param = list(self.model.parameters())    #read the param

        bias_term = np.sum(param[len(param)-1].grad.numpy() * param[len(param)-1].data.numpy())+\
                    np.sum(param[len(param)-3].grad.numpy() * param[len(param)-3].data.numpy())+\
                    np.sum(param[len(param)-5].grad.numpy() * param[len(param)-5].data.numpy())     #three FC layers


        grads_val = self.extractor.get_gradients()[-1].cpu().data.numpy()

        target = features[-1]
        target = target.cpu().data.numpy()[0, :]

        multi = np.sum(target * grads_val[0, :])

        print ("class score is %f" %(output[0][index]))
        print ("gradients*feature is %f" %(multi))
        print ("bias_term is %f" %(bias_term))
        print ("class_score-gradients*feature-bias_term= %f" %(output[0][index]-multi-bias_term))


def get_args():
    parser = argparse.ArgumentParser()
    parser.add_argument('--use-cuda', action='store_true', default=False,
                        help='Use NVIDIA GPU acceleration')
    parser.add_argument('--image-path', type=str, default='./examples/both.png',
                        help='Input image path')
    parser.add_argument('--target-index', type=int, default= -1,
                        help='class of interest')
    args = parser.parse_args()
    args.use_cuda = args.use_cuda and torch.cuda.is_available()
    if args.use_cuda:
        print("Using GPU for acceleration")
    else:
        print("Using CPU for computation")

    return args


if __name__ == '__main__':
    """ python grad_cam.py <path_to_image>
    class_score-gradients*feature-bias_term """

    args = get_args()

    # Can work with any model, but it assumes that the model has a
    # feature method, and a classifier method,
    # as in the VGG models in torchvision.
    grad_cam = GradCam(model=models.vgg16(pretrained=True), \
                       target_layer_names=["30"], use_cuda=args.use_cuda)
    print (models.vgg16(pretrained=True))
    img = cv2.imread(args.image_path, 1)
    img = np.float32(cv2.resize(img, (224, 224))) / 255
    input = preprocess_image(img)
    
    target_index = args.target_index
    grad_cam(input, target_index)