def split_image(image_fname, output_dir, tile_size, tile_step, image_margin):
os.makedirs(output_dir, exist_ok=True)
image = read_image_as_is(image_fname)
image_id = id_from_fname(image_fname)
slicer = ImageSlicer(image.shape, tile_size, tile_step, image_margin)
tiles = slicer.split(image)
fnames = []
for i, tile in enumerate(tiles):
output_fname = os.path.join(output_dir, f"{image_id}_tile_{i}.png")
cv2.imwrite(output_fname, tile)
fnames.append(output_fname)
return fnames
def __getitem__(self, i):
# read data
image = fs.read_rgb_image(self.images_fps[i])
mask = fs.read_image_as_is(self.masks_fps[i])
assert mask.max() < len(CLASSES)
# apply augmentations
sample = self.transform(image=image, mask=mask)
image, mask = sample['image'], sample['mask']
return {
"image_id": id_from_fname(self.images_fps[i]),
"features": tensor_from_rgb_image(image),
"targets": torch.from_numpy(mask).long()
}
def read_inria_mask(fname):
mask = read_image_as_is(fname)
return (mask > 0).astype(np.uint8)
def read_inria_mask_with_pseudolabel(fname):
mask = fs.read_image_as_is(fname).astype(np.float32) / 255.0
return mask
def read_inria_mask(fname):
mask = fs.read_image_as_is(fname)
if mask is None:
raise IOError("Cannot read " + fname)
cv2.threshold(mask, thresh=0, maxval=1, type=cv2.THRESH_BINARY, dst=mask)
return mask
def read_inria_mask_with_pseudolabel(fname):
mask = fs.read_image_as_is(fname)
mask[mask > UNLABELED_SAMPLE] = 1
return mask
