def preprocess_image(img_path, img_size):
img = io.imread(img_path)
if np.max(img.shape[:2]) != img_size:
# print('Resizing so the max image size is %d..' % img_size)
scale = (float(img_size) / np.max(img.shape[:2]))
else:
scale = 1.0 #scaling_factor
center = np.round(np.array(img.shape[:2]) / 2).astype(int)
# image center in (x,y)
center = center[::-1]
crop, proc_param = img_util.scale_and_crop(img, scale, center, img_size)
# import ipdb; ipdb.set_trace()
# Normalize image to [-1, 1]
# plt.imshow(crop/255.0)
# plt.show()
crop = 2 * ((crop / 255.) - 0.5)
return crop, proc_param, img
def __call__(self, sample):
imgs_to_be_processed = sample['images']
imgs_to_be_returned = []
faceposes_to_be_returned = []
img_shapes = []
scales = [] # the 224 / the actual size of img, 2 x 1
centers = [] # the center of the picture
for img_facepos in imgs_to_be_processed:
img = img_facepos['image']
if np.max(img.shape[:2]) != self.config_img_size:
# print('Resizing so the max image size is %d..' % self.config_img_size)
scale = (float(self.config_img_size) / np.max(img.shape[:2]))
else:
scale = 1.0 #scaling_factor
center = np.round(np.array(img.shape[:2]) / 2).astype(int)
# image center in (x,y)
center = center[::-1]
crop, proc_param = img_util.scale_and_crop(img, scale, center,
self.config_img_size)
# import ipdb; ipdb.set_trace()
# Normalize image to [-1, 1]
# plt.imshow(crop/255.0)
# plt.show()
crop = 2 * ((crop / 255.) - 0.5)
single_facepos = img_facepos['openpose']
# single_facepos *= scale
faceposes_to_be_returned.append(single_facepos)
imgs_to_be_returned.append(crop)
img_shapes.append(
np.array([[float(img.shape[0]), 0.], [0.,
float(img.shape[1])]]))
scales.append(scale)
centers.append(center)
return {
'images': imgs_to_be_returned,
'faceposes': faceposes_to_be_returned,
'shape': img_shapes,
'scale': scales,
'centers': centers
}
# read images and scale
#input_img_path = "./training_set/NoW_Dataset/final_release_version/iphone_pictures/FaMoS_180424_03335_TA/multiview_neutral/IMG_0101.jpg"
#input_img_path = "./training_set/NoW_Dataset/final_release_version/iphone_pictures/FaMoS_180704_03355_TA/multiview_expressions/IMG_1948.jpg"
input_img_path = "./training_set/NoW_Dataset/final_release_version/iphone_pictures/FaMoS_180427_03338_TA/multiview_expressions/IMG_0230.jpg"
#input_img_path = "./training_set/NoW_Dataset/final_release_version/iphone_pictures/FaMoS_180502_00145_TA/multiview_expressions/IMG_0407.jpg"
openpose = np.load(input_img_path.replace("iphone_pictures", "openpose").replace("jpg", "npy"), allow_pickle=True, encoding='latin1')
img = io.imread(input_img_path)
if np.max(img.shape[:2]) != config_img_size:
# print('Resizing so the max image size is %d..' % self.config_img_size)
scale = (float(config_img_size) / np.max(img.shape[:2]))
else:
scale = 1.0#scaling_factor
center = np.round(np.array(img.shape[:2]) / 2).astype(int)
# image center in (x,y)
center = center[::-1]
crop, proc_param = img_util.scale_and_crop(
img, scale, center, config_img_size)
print(proc_param)
#exit(0)
crop = torch.tensor(crop)
crop = crop.permute(2, 0, 1)
crop = crop[None, :, :, :].float().cuda()
# print(crop)
# build model
resnet50 = torch.load("./good_resnet50.pkl")
resnet50.cuda()
resnet50.fc = Identity()
# print(resnet50)
regression = torch.load("./good_model.pkl")
regression.cuda()
