#get the target images
targets_cnt[target_ind][1] += 1
target_name = id_to_name[target_ind]
target_data = []
#get one image for each target type
for t_type,_ in enumerate(target_images[target_name]):
#pick one image of this type
img_ind = np.random.choice(np.arange(
len(target_images[target_name][t_type])))
if preload_target_images:
target_img = target_images[target_name][t_type][img_ind]
else:
target_img = cv2.imread(target_images[target_name][t_type][img_ind])
if np.random.rand() > .9 and vary_images:
target_img = vary_image(target_img)
#subtract means, give batch dimension, add to list
if use_torch_vgg:
target_img = ((target_img/255.0) - [0.485, 0.456, 0.406])/[0.229, 0.224, 0.225]
else:
target_img = target_img - means
#target_img = np.expand_dims(target_img,axis=0)
batch_target_data.append(target_img)
batch_im_data.append(im_data)
#batch_gt_boxes.append(gt_boxes)
batch_gt_boxes.extend(gt_boxes)
target_path = '/net/bvisionserver3/playpen10/ammirato/Data/instance_detection_targets/AVD_BB_exact_few_and_other_BB_gen_160/'
dest_path = '/net/bvisionserver3/playpen10/ammirato/Data/instance_detection_targets/AVD_BB_exact_few_and_other_BB_gen_160_varied/'
target_types = ['target_0', 'target_1']
for t_type in target_types:
b_path = os.path.join(target_path, t_type)
dp_path = os.path.join(dest_path, t_type)
img_names = os.listdir(b_path)
for img_name in img_names:
img = cv2.imread(os.path.join(b_path, img_name))
sep_ind = img_name.rfind('_')
ext_ind = img_name.rfind('.')
obj_name = img_name[:sep_ind]
obj_index = int(img_name[sep_ind + 1:ext_ind])
cv2.imwrite(os.path.join(dp_path, img_name), img)
for il in range(1, 101):
new_img = vary_image(img)
new_index = str(obj_index + il + 100)
new_name = obj_name + '_' + new_index + '.jpg'
cv2.imwrite(os.path.join(dp_path, new_name), new_img)
