| |
| |
| |
| |
|
|
| """ Misc helper functions """ |
|
|
| import subprocess |
|
|
| import cv2 |
| import numpy as np |
| import torch |
| import torchvision.transforms as transforms |
|
|
|
|
| class NormalizeInverse(transforms.Normalize): |
| |
|
|
| def __init__(self, mean, std): |
| mean = torch.as_tensor(mean) |
| std = torch.as_tensor(std) |
| std_inv = 1 / (std + 1e-7) |
| mean_inv = -mean * std_inv |
| super(NormalizeInverse, self).__init__(mean=mean_inv, std=std_inv) |
|
|
| def __call__(self, tensor): |
| return super(NormalizeInverse, self).__call__(tensor.clone()) |
|
|
|
|
| def create_folder(folder_name): |
| try: |
| subprocess.call(["mkdir", "-p", folder_name]) |
| except OSError: |
| None |
|
|
|
|
| def save_saliency_map(image, saliency_map, filename): |
| """ |
| Save saliency map on image. |
| |
| Args: |
| image: Tensor of size (3,H,W) |
| saliency_map: Tensor of size (1,H,W) |
| filename: string with complete path and file extension |
| |
| """ |
|
|
| image = image.data.cpu().numpy() |
| saliency_map = saliency_map.data.cpu().numpy() |
|
|
| saliency_map = saliency_map - saliency_map.min() |
| saliency_map = saliency_map / saliency_map.max() |
| saliency_map = saliency_map.clip(0, 1) |
|
|
| saliency_map = np.uint8(saliency_map * 255).transpose(1, 2, 0) |
| saliency_map = cv2.resize(saliency_map, (224, 224)) |
|
|
| image = np.uint8(image * 255).transpose(1, 2, 0) |
| image = cv2.resize(image, (224, 224)) |
|
|
| |
| color_heatmap = cv2.applyColorMap(saliency_map, cv2.COLORMAP_JET) |
|
|
| |
| img_with_heatmap = np.float32(color_heatmap) + np.float32(image) |
| img_with_heatmap = img_with_heatmap / np.max(img_with_heatmap) |
|
|
| cv2.imwrite(filename, np.uint8(255 * img_with_heatmap)) |
|
|
