Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

feat: Added box coord clipping and target validation for detection tasks #355

Merged
merged 3 commits into from
Jul 6, 2021
Merged

feat: Added box coord clipping and target validation for detection tasks #355

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
e40725b
feat: Added box clipping to detection set
fg-mindee Jul 6, 2021
4af53e2
feat: Added box relative coords check
fg-mindee Jul 6, 2021
d9ab964
test: Added unittest for target checks
fg-mindee Jul 6, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion doctr/datasets/detection.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,4 +71,4 @@ def __getitem__(
boxes[..., [0, 2]] /= w
boxes[..., [1, 3]] /= h

return img, dict(boxes=boxes, flags=target['flags'])
return img, dict(boxes=boxes.clip(0, 1), flags=target['flags'])
5 changes: 5 additions & 0 deletions doctr/models/detection/differentiable_binarization/base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -254,6 +254,11 @@ def compute_target(
output_shape: Tuple[int, int, int],
) -> Tuple[np.ndarray, np.ndarray, np.ndarray, np.ndarray]:

if any(t['boxes'].dtype != np.float32 for t in target):
raise AssertionError("the 'boxes' entry of the target is expected to have dtype 'np.float32'.")
if any(np.any((t['boxes'][:, :4] > 1) | (t['boxes'][:, :4] < 0)) for t in target):
raise ValueError("the 'boxes' entry of the target is expected to take values between 0 & 1.")

seg_target = np.zeros(output_shape, dtype=np.uint8)
seg_mask = np.ones(output_shape, dtype=bool)
thresh_target = np.zeros(output_shape, dtype=np.uint8)
Expand Down
5 changes: 5 additions & 0 deletions doctr/models/detection/linknet/base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -109,6 +109,11 @@ def compute_target(
output_shape: Tuple[int, int, int],
) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:

if any(t['boxes'].dtype != np.float32 for t in target):
raise AssertionError("the 'boxes' entry of the target is expected to have dtype 'np.float32'.")
if any(np.any((t['boxes'][:, :4] > 1) | (t['boxes'][:, :4] < 0)) for t in target):
raise ValueError("the 'boxes' entry of the target is expected to take values between 0 & 1.")

if self.rotated_bbox:
seg_target = np.zeros(output_shape, dtype=np.uint8)
else:
Expand Down
14 changes: 14 additions & 0 deletions test/tensorflow/test_models_detection_tf.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,6 +38,20 @@ def test_detection_models(arch_name, input_shape, output_size, out_prob):
assert np.all(boxes[:, :4] >= 0) and np.all(boxes[:, :4] <= 1)
# Check loss
assert isinstance(out['loss'], tf.Tensor)
# Target checks
target = [
dict(boxes=np.array([[0, 0, 1, 1]], dtype=np.uint8), flags=[True, False]),
dict(boxes=np.array([[0, 0, 1, 1]], dtype=np.uint8), flags=[True, False])
]
with pytest.raises(AssertionError):
out = model(input_tensor, target, training=True)

target = [
dict(boxes=np.array([[0, 0, 1.5, 1.5]], dtype=np.float32), flags=[True, False]),
dict(boxes=np.array([[-.2, -.3, 1, 1]], dtype=np.float32), flags=[True, False])
]
with pytest.raises(ValueError):
out = model(input_tensor, target, training=True)


@pytest.fixture(scope="session")
Expand Down