def __getitem__(self, index): # TODO
img_dir = self._getInputPath(index)
if self.args.template_debug: # if debugging the template, don't read an image
img = self.images[index]
else:
img = io.imread(img_dir, as_gray=self.args.grayscale).astype(
np.float32) / 255.0
h, w, c = img.shape
crop_h, crop_w = self.args.crop_h, self.args.crop_w
if self.args.rescale:
sc_h = np.random.randint(crop_h, h)
sc_w = np.random.randint(crop_w, w)
img = transforms.rescale(img, sc_h, sc_w)
if self.args.crop:
img = transforms.random_crop(img, [crop_h, crop_w])
if self.args.color_aug:
img = (img * np.random.uniform(1, 3)).clip(0, 2)
if self.args.noise_aug:
img = transforms.random_noise_aug(img, self.args.noise)
if self.args.template_debug:
label = (index > 60) * 1.0
else:
pass # FIXME
item = {'img': torch.from_numpy(img).float(), 'label': label} # TODO
return item
def transform(self, image, target):
if self.training:
image, target = tr_custom.random_horizontal_flip([image, target])
image, target = tr_custom.random_rotation([image, target], angle=15)
image, target = tr_custom.random_crop([image, target], scale=(0.8, 1.0),
aspect_ratio=self.aspect_ratio)
else:
image, target = tr_custom.center_crop([image, target], aspect_ratio=self.aspect_ratio)
image_fullres, target = [tr.resize(img, self.fullres) for img in [image, target]]
image_lowres = tr.resize(image_fullres, self.lowres, interpolation=Image.NEAREST)
image_lowres, image_fullres, target = [tr.to_tensor(img) for img in [image_lowres, image_fullres, target]]
return image_lowres, image_fullres, target
def transform_train(img, boxes, labels):
img = random_distort(img)
# if random.random() < 0.5:
# img, boxes = random_paste(img, boxes, max_ratio=4, fill=(123, 116, 103))
img, boxes, labels = random_crop(img, boxes, labels)
img, boxes = resize(img, boxes, size=(img_size,img_size), random_interpolation=True)
img, boxes = random_flip(img, boxes)
img = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.4140, 0.4265, 0.4172), (0.2646, 0.2683, 0.2751))
])(img)
boxes, labels = box_coder.encode(boxes, labels)
return img, boxes, labels