def test_resize_transform(self):
input_shapes = [(100, 100), (100, 100, 1), (100, 100, 3)]
output_shapes = [(200, 200), (200, 200, 1), (200, 200, 3)]
for in_shape, out_shape in zip(input_shapes, output_shapes):
in_img = np.random.randint(0, 255, size=in_shape, dtype=np.uint8)
tfm = T.ResizeTransform(in_shape[0], in_shape[1], out_shape[0], out_shape[1])
out_img = tfm.apply_image(in_img)
self.assertEqual(out_img.shape, out_shape)
def run_img(img):
logger.debug('img.shape: {}'.format(img.shape))
img = img[:, :, ::-1]
if self.fix_ratio:
transform = T.ResizeTransform(*img.shape[:2], *[self.cfg.INPUT.MIN_SIZE_TEST]*2, Image.BILINEAR)
else:
transform = self.transform_gen.get_transform(img)
img = apply_image(transform, img)
input_y, input_x = img.shape[:2]
if _boxes is None:
boxes = [[0, 0, input_x, input_y]]
else:
boxes = np.asarray(_boxes).reshape(-1, 2)
boxes = transform.apply_coords(boxes)
boxes = boxes.reshape(-1, 4)
with torch.no_grad():
img = torch.as_tensor(img.astype('float32').transpose(2, 0, 1))
model = self.model
images = model.preprocess_image([{'image': img}])
features = model.backbone(images.tensor)
features = [features[f] for f in model.roi_heads.in_features]
x = model.roi_heads.pooler(features, [Boxes(torch.tensor(boxes, dtype=torch.float32, device=model.device))])
x = model.roi_heads.res5[:res5_i](x)
return x.cpu().numpy().transpose(0, 2, 3, 1)
def get_transform(self, image, sem_seg):
return T.ResizeTransform(input_shape[0], input_shape[1],
output_shape[0], output_shape[1])
def get_transform(self, image):
old_h, old_w = image.shape[:2]
new_h, new_w = int(old_h * np.random.rand()), int(old_w * 1.5)
return T.ResizeTransform(old_h, old_w, new_h, new_w)
