def test_Image(self, debug=True):
"""
Usage:
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
export TIME_STR=1
export PYTHONPATH=./exp:./stylegan2-pytorch:./
python -c "from exp.tests.test_styleganv2 import Testing_stylegan2;\
Testing_stylegan2().test_train_ffhq_128()"
:return:
"""
if 'CUDA_VISIBLE_DEVICES' not in os.environ:
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
if 'TIME_STR' not in os.environ:
os.environ['TIME_STR'] = '0' if utils.is_debugging() else '0'
from template_lib.v2.config_cfgnode.argparser import \
(get_command_and_outdir, setup_outdir_and_yaml, get_append_cmd_str, start_cmd_run)
tl_opts = ' '.join(sys.argv[sys.argv.index('--tl_opts') +
1:]) if '--tl_opts' in sys.argv else ''
print(f'tl_opts:\n {tl_opts}')
command, outdir = get_command_and_outdir(
self, func_name=sys._getframe().f_code.co_name, file=__file__)
argv_str = f"""
--tl_config_file none
--tl_command none
--tl_outdir {outdir}
--tl_opts {tl_opts}
"""
args, cfg = setup_outdir_and_yaml(argv_str, return_cfg=True)
n_gpus = len(os.environ['CUDA_VISIBLE_DEVICES'].split(','))
import mmcv
import numpy as np
import matplotlib.pyplot as plt
img_path = "template_lib/datasets/images/zebra_GT_target_origin.png"
img = mmcv.imread(img_path)
img_gray = mmcv.imread(img_path, flag='grayscale')
img_ = mmcv.imread(img) # nothing will happen, img_ = img
mmcv.imwrite(img, f'{args.tl_outdir}/out.png')
# mmcv.imshow(img)
fig, axes = plt.subplots(2, 1)
img = mmcv.bgr2rgb(img)
axes[0].imshow(img)
# plt.imshow(img)
# plt.show()
# ret = mmcv.imresize(img, (1000, 600), return_scale=True)
# ret = mmcv.imrescale(img, (1000, 800))
bboxes = np.array([10, 10, 100, 120])
patch = mmcv.imcrop(img, bboxes)
axes[1].imshow(patch)
fig.show()
pass
def __call__(self, results):
bbox_xywh = bbox_xyxy2xywh(results['bbox'][None, :])[0]
length = bbox_xywh[2:].max()
length = length * np.random.uniform(*self.scale)
x = bbox_xywh[0] + np.random.uniform(-self.shift, self.shift) * length
y = bbox_xywh[1] + np.random.uniform(-self.shift, self.shift) * length
w, h = length, length * np.random.uniform(*self.ratio)
bbox_xyxy = bbox_xywh2xyxy(np.array([[x, y, w, h]]))[0]
bbox_xyxy = bbox_xyxy.clip(min=0)
bbox_xyxy[2] = min(bbox_xyxy[2], results['width'][0])
bbox_xyxy[3] = min(bbox_xyxy[3], results['height'][0])
bbox_xyxy = bbox_xyxy.astype(np.int32)
for i in range(len(results['video'])):
video = results['video'][i].transpose(1, 2, 3, 0)
num_frames = video.shape[-1]
video = video.reshape(video.shape[0], video.shape[1], -1)
video = mmcv.imcrop(video, bbox_xyxy)
video = mmcv.imresize(video, self.size)
results['video'][i] = video.reshape(video.shape[0], video.shape[1],
-1, num_frames)
results['video'][i] = results['video'][i].transpose(3, 0, 1, 2)
results['width'][i], results['height'][i] = video.shape[
1], video.shape[0]
return results
def __call__(self, results):
for key in results.get('img_fields', ['img']):
img = results[key]
if self.efficientnet_style:
get_params_func = self.get_params_efficientnet_style
get_params_args = dict(
img=img,
size=self.size,
scale=self.scale,
ratio=self.ratio,
max_attempts=self.max_attempts,
min_covered=self.min_covered,
crop_padding=self.crop_padding)
else:
get_params_func = self.get_params
get_params_args = dict(
img=img,
scale=self.scale,
ratio=self.ratio,
max_attempts=self.max_attempts)
ymin, xmin, ymax, xmax = get_params_func(**get_params_args)
img = mmcv.imcrop(img, bboxes=np.array([xmin, ymin, xmax, ymax]))
results[key] = mmcv.imresize(
img,
tuple(self.size[::-1]),
interpolation=self.interpolation,
backend=self.backend)
return results
def __call__(self, results):
"""Call function.
Args:
results (dict): A dict containing the necessary information and
data for augmentation.
Returns:
dict: A dict containing the processed data and information.
"""
for key in self.keys:
img = results[key]
img_height, img_width = img.shape[:2]
crop_size = min(img_height, img_width)
y1 = 0 if img_height == crop_size else \
int(round(img_height - crop_size) / 2)
x1 = 0 if img_width == crop_size else \
int(round(img_width - crop_size) / 2)
y2 = y1 + crop_size - 1
x2 = x1 + crop_size - 1
img = mmcv.imcrop(img, bboxes=np.array([x1, y1, x2, y2]))
results[key] = img
return results
def __call__(self, results):
crop_height, crop_width = self.crop_size[0], self.crop_size[1]
for key in results.get('img_fields', ['img']):
img = results[key]
# img.shape has length 2 for grayscale, length 3 for color
img_height, img_width = img.shape[:2]
# https://github.com/tensorflow/tpu/blob/master/models/official/efficientnet/preprocessing.py#L118 # noqa
if self.efficientnet_style:
img_short = min(img_height, img_width)
crop_height = crop_height / (crop_height +
self.crop_padding) * img_short
crop_width = crop_width / (crop_width +
self.crop_padding) * img_short
y1 = max(0, int(round((img_height - crop_height) / 2.)))
x1 = max(0, int(round((img_width - crop_width) / 2.)))
y2 = min(img_height, y1 + crop_height) - 1
x2 = min(img_width, x1 + crop_width) - 1
# crop the image
img = mmcv.imcrop(img, bboxes=np.array([x1, y1, x2, y2]))
if self.efficientnet_style:
img = mmcv.imresize(
img,
tuple(self.crop_size[::-1]),
interpolation=self.interpolation,
backend=self.backend)
img_shape = img.shape
results[key] = img
results['img_shape'] = img_shape
return results
def __call__(self, results):
"""
Args:
img (ndarray): Image to be cropped.
"""
for key in results.get('img_fields', ['img']):
img = results[key]
if self.padding is not None:
img = mmcv.impad(img,
padding=self.padding,
pad_val=self.pad_val)
# pad the height if needed
if self.pad_if_needed and img.shape[0] < self.size[0]:
img = mmcv.impad(img,
padding=(0, self.size[0] - img.shape[0], 0,
self.size[0] - img.shape[0]),
pad_val=self.pad_val,
padding_mode=self.padding_mode)
# pad the width if needed
if self.pad_if_needed and img.shape[1] < self.size[1]:
img = mmcv.impad(img,
padding=(self.size[1] - img.shape[1], 0,
self.size[1] - img.shape[1], 0),
pad_val=self.pad_val,
padding_mode=self.padding_mode)
xmin, ymin, height, width = self.get_params(img, self.size)
results[key] = mmcv.imcrop(
img,
np.array([ymin, xmin, ymin + width - 1, xmin + height - 1]))
return results
def __call__(self, results):
img_group = results['img_group']
img_h, img_w = img_group[0].shape[:2]
crop_w, crop_h = self.crop_size
offsets = MultiScaleCrop.fill_fix_offset(False, img_w, img_h, crop_w,
crop_h)
oversample_group = list()
for o_w, o_h in offsets:
normal_group = list()
flip_group = list()
for i, img in enumerate(img_group):
crop = mmcv.imcrop(
img,
np.array([o_w, o_h, o_w + crop_w - 1, o_h + crop_h - 1]))
normal_group.append(crop)
flip_crop = mmcv.imflip(crop)
if results['modality'] == 'Flow' and i % 2 == 0:
flip_group.append(mmcv.iminvert(flip_crop))
else:
flip_group.append(flip_crop)
oversample_group.extend(normal_group)
oversample_group.extend(flip_group)
results['img_group'] = oversample_group
results['crop_bbox'] = None
results['img_shape'] = results['img_group'][0].shape
return results
def __call__(self, img_group, is_flow=False):
image_h = img_group[0].shape[0]
image_w = img_group[0].shape[1]
crop_w, crop_h = self.crop_size
offsets = GroupMultiScaleCrop.fill_fix_offset(
False, image_w, image_h, crop_w, crop_h)
oversample_group = list()
for o_w, o_h in offsets:
normal_group = list()
flip_group = list()
for i, img in enumerate(img_group):
crop = mmcv.imcrop(img, np.array(
[o_w, o_h, o_w + crop_w-1, o_h + crop_h-1]))
normal_group.append(crop)
flip_crop = mmcv.imflip(crop)
if is_flow and i % 2 == 0:
flip_group.append(mmcv.iminvert(flip_crop))
else:
flip_group.append(flip_crop)
oversample_group.extend(normal_group)
oversample_group.extend(flip_group)
return oversample_group, None
def crop_bboxes(self, image_id, bboxes):
img_path = osp.join("data/PRW/frames", self.id_2_filename(image_id))
img = mmcv.imread(img_path)
crops = []
for bbox in bboxes:
xyxy = np.asarray([int(bbox[0]), int(bbox[1]), int(bbox[0] + bbox[2]), int(bbox[1] + bbox[3])])
crops.append(mmcv.imcrop(img, xyxy))
return crops
def __call__(self, img):
h, w = img.shape[:2]
assert w >= self.img_scale[0] and h >= self.img_scale[1]
x1 = w // 2 - self.img_scale[0] // 2
x2 = x1 + self.img_scale[0]
y1 = h // 2 - self.img_scale[1] // 2
y2 = x1 + self.img_scale[1]
img = mmcv.imcrop(img, np.array([x1, y1, x2, y2]))
return img
def __call__(self, img_group, is_flow=False):
h = img_group[0].shape[0]
w = img_group[0].shape[1]
tw, th = self.size
x1 = (w - tw) // 2
y1 = (h - th) // 2
box = np.array([x1, y1, x1+tw-1, y1+th-1])
return ([mmcv.imcrop(img, box) for img in img_group],
np.array([x1, y1, tw, th], dtype=np.float32))
def __getitem__(self, item):
img_id = self.img_infos[item]['id']
filename = self.img_infos[item]['filename']
#print(self.boxes.keys())
#print(img_id)
#ss_box = self.boxes[img_id][:30,:]
#ss_box = self.mesh
#print(ss_box.shape)
bbox_label = self.get_ann_info(item)
image = cv2.imread(osp.join(self.img_prefix, filename))
bboxes = bbox_label['bboxes'] #person
labeles = bbox_label['labels']
randidx = np.random.randint(0, len(labeles))
bbox = bboxes[randidx]
label = labeles[randidx] - 1
#assert label!=-1,img_id
#if label==10:
# self.count +=1
# print(self.count)
if self.crop:
#print("crop person image!!!!!")
person_img = mmcv.imcrop(image, bbox, scale=1.0)
else:
person_img = image
import os
#path_person = os.path.join(self.img_prefix,'PersonImages')
#if not os.path.exists(path_person):
# os.makedirs(path_person)
#cv2.imwrite(os.path.join(path_person,img_id+'.jpg'),person_img)
# cv2 to PIL
person_img = Image.fromarray(
cv2.cvtColor(person_img, cv2.COLOR_BGR2RGB))
#img = Image.fromarray(cv2.cvtColor(image, cv2.COLOR_BGR2RGB))
#transform
if self.train:
trn = T.Compose([
#T.RandomResizedCrop(224, (0.3, 1.0)),
#T.RandomHorizontalFlip(0.7),
# T.RandomRotation(30),
T.Resize(self.cfg.DATASET.IMAGE_SIZE),
T.ToTensor(),
T.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
])
else:
trn = T.Compose([
T.Resize(self.cfg.DATASET.IMAGE_SIZE),
T.ToTensor(),
T.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
])
person_img = trn(person_img)
#img = trn(img)
#return ss_box.astype(np.float32),person_img,label
return person_img, label
def crop_img(self, image_id):
# 将image_id做为下标
crops = []
img_path = osp.join("data/PRW/frames", self.id_2_filename(image_id))
img = mmcv.imread(img_path)
for det in self.detections:
if int(det["image_id"]) == image_id and int(det['category_id']) == 1:
bbox = det["bbox"]
xyxy = np.asarray([int(bbox[0]), int(bbox[1]), int(bbox[0] + bbox[2]), int(bbox[1] + bbox[3])])
crops.append(mmcv.imcrop(img, xyxy))
return crops
def get_frame_crop(self, det):
crops = []
if len(det) > 0:
image_id = det[0]["image_id"]
bboxes = []
img_path = osp.join("data/PRW/frame", self.id_2_filename(image_id))
img = mmcv.imread(img_path)
for d in det:
bboxes.append(d["bbox"])
crops = mmcv.imcrop(img, bboxes)
return crops
def __call__(self, img_group, is_flow=False, interpolation='bilinear'):
ret_img_group = [
mmcv.imcrop(img, self.crop_quadruple) for img in img_group
]
if self.resize:
ret_img_group = [
mmcv.imresize(img, (self.input_size[0], self.input_size[1]),
interpolation=interpolation)
for img in ret_img_group
]
return ret_img_group, self.crop_quadruple
def _inference_vos_single(model, img, img_refer, bbox, img_transform, device):
img = mmcv.imread(img)
# crop the object in first frame
img_refer = mmcv.imread(img_refer)
# crop the bbox
img_refer = mmcv.imcrop(img_refer, bbox)
data = _prepare_vos_data(img, img_refer, img_transform, model.cfg, device)
with torch.no_grad():
result = model(return_loss=False, rescale=True, **data)
return result
def __call__(self, results):
img_group = results['img_group']
img_h, img_w = img_group[0].shape[:2]
crop_w, crop_h = self.crop_size
# assert crop_h == img_h or crop_w == img_w
if crop_h == img_h:
w_step = (img_w - crop_w) // 2
offsets = [
(0, 0), # left
(2 * w_step, 0), # right
(w_step, 0), # middle
]
elif crop_w == img_w:
h_step = (img_h - crop_h) // 2
offsets = [
(0, 0), # top
(0, 2 * h_step), # down
(0, h_step), # middle
]
else:
w_step = (img_w - crop_w) // 4
h_step = (img_h - crop_h) // 4
offsets = list()
offsets.append((0 * w_step, 2 * h_step)) # left
offsets.append((4 * w_step, 2 * h_step)) # right
offsets.append((2 * w_step, 2 * h_step)) # center
oversample_group = list()
for o_w, o_h in offsets:
normal_group = list()
flip_group = list()
for i, img in enumerate(img_group):
crop = mmcv.imcrop(
img,
np.array([o_w, o_h, o_w + crop_w - 1, o_h + crop_h - 1]))
normal_group.append(crop)
flip_crop = mmcv.imflip(crop)
if results['modality'] == 'Flow' and i % 2 == 0:
flip_group.append(mmcv.iminvert(flip_crop))
else:
flip_group.append(flip_crop)
oversample_group.extend(normal_group)
results['img_group'] = oversample_group
results['crop_bbox'] = None
results['img_shape'] = results['img_group'][0].shape
return results
def __call__(self, results):
img_group = results['img_group']
img_h, img_w = img_group[0].shape[:2]
crop_w, crop_h = self.crop_size
x1 = (img_w - crop_w) // 2
y1 = (img_h - crop_h) // 2
box = np.array([x1, y1, x1 + crop_w - 1, y1 + crop_h - 1])
results['img_group'] = [mmcv.imcrop(img, box) for img in img_group]
results['crop_bbox'] = box
results['img_shape'] = results['img_group'][0].shape
return results
def __call__(self, img_group):
"""
Args:
clip (list of PIL Image): list of Image to be cropped and resized.
Returns:
list of PIL Image: Randomly cropped and resized image.
"""
x1, y1, th, tw = self.get_params(img_group[0], self.scale, self.ratio)
box = np.array([x1, y1, x1 + tw - 1, y1 + th - 1], dtype=np.float32)
return ([
mmcv.imresize(mmcv.imcrop(img, box), self.size)
for img in img_group
], box)
def __call__(self, img_group, is_flow=False):
image_height = img_group[0].shape[0]
image_width = img_group[0].shape[1]
crop_width, crop_height = self.size
x1 = (image_width - crop_width) // 2
y1 = (image_height - crop_height) // 2
if self.portrait_mode:
y1 = min(y1, x1)
box = np.array([x1, y1, x1 + crop_width - 1, y1 + crop_height - 1])
return ([mmcv.imcrop(img, box) for img in img_group],
np.array([x1, y1, crop_width, crop_height], dtype=np.float32))
def __call__(self, img_group, is_flow=False):
im_h = img_group[0].shape[0]
im_w = img_group[0].shape[1]
crop_w, crop_h, offset_w, offset_h = self._sample_crop_size(
(im_w, im_h))
box = np.array([offset_w, offset_h, offset_w +
crop_w - 1, offset_h + crop_h - 1])
crop_img_group = [mmcv.imcrop(img, box) for img in img_group]
ret_img_group = [mmcv.imresize(
img, (self.input_size[0], self.input_size[1]),
interpolation=self.interpolation)
for img in crop_img_group]
return (ret_img_group, np.array([offset_w, offset_h, crop_w, crop_h],
dtype=np.float32))
def detect_and_classify(self, img_filepath):
"""
test a single image
"""
tik = time.time()
result = {"status": 0, "data": {}}
img = mmcv.imread(img_filepath)
res = inference_detector(self.detector, img, self.cfg)
if len(res) != 0:
f_index = 0
for cat, bbox in enumerate(res):
print(cat, bbox)
if len(bbox) != 0:
# crop patches with mmcv
bboxes = np.array(bbox[:, 0:-1]).astype(
np.int
) # 0:-1 represents that the last col is confidence, and the first 4 cols are x1 y1 x2 y2
patches = mmcv.imcrop(img, bboxes)
for patch in patches:
patch = PIL.Image.fromarray(patch)
b, g, r = patch.split()
patch = Image.merge("RGB", (r, g, b))
if self.crop:
patch.save('{0}.jpg'.format(f_index))
cat_name = self.classify_from_mat(patch)
print(cat_name)
if cat_name in result['data'].keys():
result['data'][cat_name] += 1
else:
result['data'][cat_name] = 1
f_index += 1
# else:
# if "Bad" not in result['data'].keys():
# result['data']["Bad"] = 1
# else:
# result['data']["Bad"] += 1
tok = time.time()
else:
result['data'] = {"Bad": 1}
result['elapse'] = tok - tik
return result
def __call__(self, results):
crop_height, crop_width = self.crop_size[0], self.crop_size[1]
for key in results.get('img_fields', ['img']):
img = results[key]
img_height, img_width, _ = img.shape
y1 = max(0, int(round((img_height - crop_height) / 2.)))
x1 = max(0, int(round((img_width - crop_width) / 2.)))
y2 = min(img_height, y1 + crop_height) - 1
x2 = min(img_width, x1 + crop_width) - 1
# crop the image
img = mmcv.imcrop(img, bboxes=np.array([x1, y1, x2, y2]))
img_shape = img.shape
results[key] = img
results['img_shape'] = img_shape
return results
def __call__(self, results):
"""Implement data processing with center crop."""
for i, modal in enumerate(results['modality']):
height, width = results['height'][i], results['width'][i]
video = results['video'][i].transpose(1, 2, 3, 0)
num_frames = video.shape[-1]
video = video.reshape(height, width, -1)
start_h, start_w = (height - self.length) // 2, (width -
self.length) // 2
video = mmcv.imcrop(
video,
np.array((start_w, start_h, start_w + self.length - 1,
start_h + self.length - 1)))
results['video'][i] = video.reshape(self.length, self.length, -1,
num_frames).transpose(
3, 0, 1, 2)
results['width'][i], results['height'][
i] = self.length, self.length
return results
def __call__(self, results):
img_group = results['img_group']
img_h, img_w = img_group[0].shape[:2]
(crop_w, crop_h), (offset_w, offset_h) = self._sample_crop_size(
(img_w, img_h))
box = np.array(
[offset_w, offset_h, offset_w + crop_w - 1, offset_h + crop_h - 1])
crop_img_group = [mmcv.imcrop(img, box) for img in img_group]
ret_img_group = [
mmcv.imresize(img, (self.input_size[0], self.input_size[1]),
interpolation=self.interpolation)
for img in crop_img_group
]
results['crop_bbox'] = box
results['img_group'] = ret_img_group
results['img_shape'] = ret_img_group[0].shape
results['scales'] = self.scales
return results
def __call__(self, results):
"""
Args:
clip (list of PIL Image): list of Image to be cropped and resized.
Returns:
list of PIL Image: Randomly cropped and resized image.
"""
img_group = results['img_group']
(x1, y1), (crop_h, crop_w) = self.get_params(img_group[0], self.scale,
self.ratio)
box = np.array([x1, y1, x1 + crop_w - 1, y1 + crop_h - 1],
dtype=np.float32)
results['img_group'] = [
mmcv.imresize(mmcv.imcrop(img, box), self.input_size)
for img in img_group
]
results['crop_bbox'] = box
results['img_shape'] = results['img_group'][0].shape
return results
def __call__(self, img_group, is_flow=False):
image_h = img_group[0].shape[0]
image_w = img_group[0].shape[1]
crop_w, crop_h = self.crop_size
assert crop_h == image_h or crop_w == image_w
if crop_h == image_h:
w_step = (image_w - crop_w) // 2
offsets = list()
offsets.append((0, 0)) # left
offsets.append((2 * w_step, 0)) # right
offsets.append((w_step, 0)) # middle
elif crop_w == image_w:
h_step = (image_h - crop_h) // 2
offsets = list()
offsets.append((0, 0)) # top
offsets.append((0, 2 * h_step)) # down
offsets.append((0, h_step)) # middle
oversample_group = list()
for o_w, o_h in offsets:
normal_group = list()
flip_group = list()
for i, img in enumerate(img_group):
crop = mmcv.imcrop(
img,
np.array([o_w, o_h, o_w + crop_w - 1, o_h + crop_h - 1]))
normal_group.append(crop)
flip_crop = mmcv.imflip(crop)
if is_flow and i % 2 == 0:
flip_group.append(mmcv.iminvert(flip_crop))
else:
flip_group.append(flip_crop)
oversample_group.extend(normal_group)
# oversample_group.extend(flip_group)
return oversample_group, None
def process_image(image, augment, square=False):
if square:
iH, iW = 240, 240
H, W = 224, 224
else:
iH, iW = 240, 480
H, W = 224, 448
image = mmcv.imresize(image, (iW, iH))
h, w, _ = image.shape
w_offset = w - W
h_offset = h - H
left = random.randrange(0, w_offset + 1)
upper = random.randrange(0, h_offset + 1)
image = mmcv.imcrop(image,
np.array([left, upper, left + W - 1, upper + H - 1]))
if augment:
enhancer = ImageEnhance.Brightness(Image.fromarray(image))
image = enhancer.enhance(random.random() * 0.6 + 0.7)
enhancer = ImageEnhance.Color(image)
image = enhancer.enhance(random.random() * 0.6 + 0.7)
return image
def __call__(self, results):
"""
Args:
img (ndarray): Image to be cropped and resized.
Returns:
ndarray: Randomly cropped and resized image.
"""
for key in results.get('img_fields', ['img']):
img = results[key]
xmin, ymin, target_height, target_width = self.get_params(
img, self.scale, self.ratio)
img = mmcv.imcrop(
img,
np.array([
ymin, xmin, ymin + target_width - 1,
xmin + target_height - 1
]))
results[key] = mmcv.imresize(img,
tuple(self.size[::-1]),
interpolation=self.interpolation)
return results
def __getitem__(self, item):
img_id = self.img_infos[item]['id']
filename = self.img_infos[item]['filename']
bbox_label = self.get_ann_info(item)
image = cv2.imread(osp.join(self.img_prefix,filename))
bboxes = bbox_label['bboxes']
labeles = bbox_label['labels']
randidx = np.random.randint(0,len(labeles))
bbox = bboxes[randidx]
label = labeles[randidx] - 1
#assert label!=-1,img_id
#if label==10:
# self.count +=1
# print(self.count)
if self.cfg.DATASET.CROP_BBOX:
#print("crop person image!!!!!")
person_img = mmcv.imcrop(image,bbox,scale=1.0)
else:
person_img = image
# cv2 to PIL
person_img = Image.fromarray(cv2.cvtColor(person_img, cv2.COLOR_BGR2RGB))
#transform
#if self.train:
trn = T.Compose([T.Resize(self.cfg.DATASET.IMAGE_SIZE),
T.RandomResizedCrop(224,(0.5,1.0)),
T.RandomHorizontalFlip(0.7),
T.RandomRotation((0,0.5)),
T.ToTensor(),
T.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.5,0.5,0.5])])
#else:
trn = T.Compose([T.Resize(self.cfg.DATASET.IMAGE_SIZE),
T.ToTensor(),
T.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.5, 0.5, 0.5])])
person_img = trn(person_img)
return person_img,label
