def prepare_single(img, frame_id, scale, flip, proposal=None):
ori_shape = img.shape
_img, img_shape, pad_shape, scale_factor = self.img_transform(
img, scale, flip, keep_ratio=self.resize_keep_ratio)
_img = to_tensor(_img)
ori_shape = img.shape
_img_meta = dict(ori_shape=ori_shape,
img_shape=img_shape,
pad_shape=pad_shape,
is_first=(frame_id == 0),
video_id=vid,
frame_id=frame_id,
scale_factor=scale_factor,
flip=flip)
if proposal is not None:
if proposal.shape[1] == 5:
score = proposal[:, 4, None]
proposal = proposal[:, :4]
else:
score = None
_proposal = self.bbox_transform(proposal, img_shape,
scale_factor, flip)
_proposal = np.hstack([_proposal, score
]) if score is not None else _proposal
_proposal = to_tensor(_proposal)
else:
_proposal = None
return _img, _img_meta, _proposal
def _prepare_data_3d_2scales(img_np, img_np_2, img_transform, cfg, device):
ori_shape = (img_np.shape[0], img_np.shape[1], 3)
ori_shape_2 = (img_np_2.shape[0], img_np_2.shape[1], 3)
total_num_slices = img_np.shape[2]
total_num_slices_2 = img_np_2.shape[2]
# first image
imgs = []
for cur_slice in range(total_num_slices):
img = img_np[:,:,cur_slice]
img = Image.fromarray(img).convert('RGB')
img = np.array(img)
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
img, img_shape, pad_shape, scale_factor = img_transform(
img,
scale=cfg.data.test.img_scale,
keep_ratio=cfg.data.test.get('resize_keep_ratio', True))
imgs.append(img)
imgs = to_tensor(np.array(imgs)).to(device).unsqueeze(0)
img_meta = [
dict(
ori_shape=ori_shape,
img_shape=(*img_shape, total_num_slices),
pad_shape=(*pad_shape, total_num_slices),
# scale_factor=1.0 / (img_np_2.shape[0] / img_np.shape[0]), # scale up to 1.5x
scale_factor=1.0, # scale down 1.0x
flip=False)
]
imgs = imgs.permute(0, 2, 1, 3, 4)
# second image
imgs_2 = []
for cur_slice in range(total_num_slices_2):
img = img_np_2[:,:,cur_slice]
img = Image.fromarray(img).convert('RGB')
img = np.array(img)
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
img, img_shape, pad_shape, scale_factor = img_transform(
img,
scale=cfg.data2_2scales.test.img_scale,
keep_ratio=cfg.data.test.get('resize_keep_ratio', True))
imgs_2.append(img)
imgs_2 = to_tensor(np.array(imgs_2)).to(device).unsqueeze(0)
img_meta_2 = [
dict(
ori_shape=ori_shape_2,
img_shape=(*img_shape, total_num_slices_2),
pad_shape=(*pad_shape, total_num_slices_2),
# scale_factor=scale_factor, # scale up to 1.5x
scale_factor=1.5, # scale down 1.0x
flip=False)
]
imgs_2 = imgs_2.permute(0, 2, 1, 3, 4)
assert imgs.shape[1] == 3 # make sure channel size is 3
assert imgs_2.shape[1] == 3
return dict(imgs=imgs, img_meta=[img_meta], imgs_2=imgs_2, img_meta_2=[img_meta_2])
def parepare_img(self, img):
if self.new_image:
cfg = self.model.cfg
img_transform = ImageTransform(
size_divisor=cfg.data.test.size_divisor, **cfg.img_norm_cfg)
device = next(self.model.parameters()).device
img = mmcv.imread(img)
ori_shape = img.shape
img, img_shape, pad_shape, scale_factor = img_transform(
img,
scale=cfg.data.test.img_scale,
keep_ratio=cfg.data.test.get('resize_keep_ratio', True))
img = to_tensor(img).to(device).unsqueeze(0)
img_meta = [
dict(
ori_shape=ori_shape,
img_shape=img_shape,
pad_shape=pad_shape,
scale_factor=scale_factor,
flip=False)
]
self.img=img
self.img_meta=img_meta
def _prepare_data_3d(img_np, img_transform, cfg, device):
ori_shape = (img_np.shape[0], img_np.shape[1], 3)
total_num_slices = img_np.shape[2]
imgs = []
for cur_slice in range(total_num_slices):
img = img_np[:,:,cur_slice]
img = Image.fromarray(img).convert('RGB')
img = np.array(img)
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
img, img_shape, pad_shape, scale_factor = img_transform(
img,
scale=cfg.data.test.img_scale,
keep_ratio=cfg.data.test.get('resize_keep_ratio', True))
imgs.append(img)
imgs = to_tensor(np.array(imgs)).to(device).unsqueeze(0)
img_meta = [
dict(
ori_shape=ori_shape,
img_shape=(*img_shape, total_num_slices),
pad_shape=(*pad_shape, total_num_slices),
scale_factor=scale_factor,
flip=False)
]
imgs = imgs.permute(0, 2, 1, 3, 4)
assert imgs.shape[1] == 3 # make sure channel size is 3
return dict(imgs=imgs, img_meta=[img_meta])
def _prepare_vos_data(img, img_refer, img_transform, cfg, device):
ori_shape = img.shape
img, img_shape, pad_shape, scale_factor = img_transform(
img,
scale=cfg.data.test.img_scale,
keep_ratio=cfg.data.test.get('resize_keep_ratio', True))
img = to_tensor(img).to(device).unsqueeze(0)
img_meta = [
dict(ori_shape=ori_shape,
img_shape=img_shape,
pad_shape=pad_shape,
scale_factor=scale_factor,
flip=False)
]
img_refer = torch.unsqueeze(to_tensor(
mmcv.imresize(np.float32(img_refer),
cfg.data.test.refer_scale,
return_scale=False)).permute(2, 0, 1),
dim=0).to(device)
return dict(img=[img], img_meta=[img_meta], img_refer=[img_refer])
def _prepare_data(img, img_transform, cfg, device):
ori_shape = img.shape
img, img_shape, pad_shape, scale_factor = img_transform(
img, scale=cfg.data.test.img_scale)
img = to_tensor(img).to(device).unsqueeze(0)
img_meta = [
dict(ori_shape=ori_shape,
img_shape=img_shape,
pad_shape=pad_shape,
scale_factor=scale_factor,
flip=False)
]
return dict(img=[img], img_meta=[img_meta])
def _prepare_single_data(img, img_transform, scale, cfg, device):
ori_shape = img.shape
print(ori_shape)
img, img_shape, pad_shape, scale_factor = img_transform(
img,
scale=scale,
keep_ratio=cfg.data.test.get('resize_keep_ratio', True))
img = to_tensor(img).to(device).unsqueeze(0)
img_meta = dict(ori_shape=ori_shape,
img_shape=img_shape,
pad_shape=pad_shape,
scale_factor=scale_factor,
flip=False)
return img, img_meta
def _prepare_data(img, img_transform, cfg, device):
ori_shape = img.shape
img, img_shape, pad_shape, scale_factor = img_transform(
img,
scale=cfg.data.test.img_scale,
keep_ratio=cfg.data.test.get('resize_keep_ratio', True))
img = to_tensor(img).to(device).unsqueeze(0)
img_metas = [
dict(ori_shape=ori_shape,
img_shape=img_shape,
pad_shape=pad_shape,
scale_factor=scale_factor,
flip=False)
]
return dict(img=[img], img_metas=[img_metas])
def _prepare_data(img, img_transform, cfg, device):
ori_shape = img.shape
# 执行img_transform实例的call方法，直接调用
# 进行图片的归一化/翻转等增强
img, img_shape, pad_shape, scale_factor = img_transform(
img,
scale=cfg.data.test.img_scale,
keep_ratio=cfg.data.test.get('resize_keep_ratio', True))
img = to_tensor(img).to(device).unsqueeze(0)
# 写入img_meta相关信息
img_meta = [
dict(ori_shape=ori_shape,
img_shape=img_shape,
pad_shape=pad_shape,
scale_factor=scale_factor,
flip=False)
]
return dict(img=[img], img_meta=[img_meta])
def _prepare_data(img, img_transform, cfg, flip, trans_scale, device):
if flip: img = cv2.flip(img, 1)
ori_shape = img.shape
height, width = img.shape[:2]
transform_scale = trans_scale
img, img_shape, pad_shape, scale_factor = img_transform(
img,
scale=transform_scale,
keep_ratio=cfg.data.test.get('resize_keep_ratio', True))
img = to_tensor(img).to(device).unsqueeze(0)
img_meta = [
dict(
ori_shape=ori_shape,
img_shape=img_shape,
pad_shape=pad_shape,
scale_factor=scale_factor,
flip=False)
]
# print(img_meta)
return dict(img=[img], img_meta=[img_meta])