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

Bugfix - Object Detection IOU #817

Merged
merged 2 commits into from
Nov 8, 2024
Merged

Bugfix - Object Detection IOU #817

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
0f5602e
fixed iou computation in object detection
czaloom Nov 8, 2024
15bd4f3
suggested change
czaloom Nov 8, 2024
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
59 changes: 55 additions & 4 deletions lite/tests/object_detection/test_dataloader.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,49 @@ def test_no_data():
loader.finalize()


def test_iou_computation():

detection = Detection(
uid="uid",
groundtruths=[
BoundingBox(xmin=0, xmax=10, ymin=0, ymax=10, labels=["0"]),
BoundingBox(xmin=100, xmax=110, ymin=100, ymax=110, labels=["0"]),
BoundingBox(
xmin=1000, xmax=1100, ymin=1000, ymax=1100, labels=["0"]
),
],
predictions=[
BoundingBox(
xmin=1,
xmax=11,
ymin=1,
ymax=11,
labels=["0", "1", "2"],
scores=[0.5, 0.25, 0.25],
),
BoundingBox(
xmin=105,
xmax=116,
ymin=105,
ymax=116,
labels=["0", "1", "2"],
scores=[0.5, 0.25, 0.25],
),
],
)

loader = DataLoader()
loader.add_bounding_boxes([detection])

assert len(loader.pairs) == 1

# show that three unique IOUs exist
unique_ious = np.unique(loader.pairs[0][:, 3])
assert np.isclose(
unique_ious, np.array([0.0, 0.12755102, 0.68067227])
).all()


def test_mixed_annotations(
rect1: tuple[float, float, float, float],
rect1_rotated_5_degrees_around_origin: tuple[float, float, float, float],
Expand Down Expand Up @@ -87,7 +130,15 @@ def test_mixed_annotations(

loader = DataLoader()

for input_ in mixed_detections:
with pytest.raises(ValueError) as e:
loader.add_bounding_boxes([input_])
assert "but annotation is of type" in str(e)
for detection in mixed_detections:

# anything can be converted to a bbox
loader.add_bounding_boxes([detection])

with pytest.raises(AttributeError) as e:
loader.add_polygons([detection])
assert "no attribute 'shape'" in str(e)

with pytest.raises(AttributeError) as e:
loader.add_bitmasks([detection])
assert "no attribute 'mask'" in str(e)
24 changes: 0 additions & 24 deletions lite/valor_lite/object_detection/annotation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -142,18 +142,6 @@ def extrema(self) -> tuple[float, float, float, float]:
xmin, ymin, xmax, ymax = self.shape.bounds
return (xmin, xmax, ymin, ymax)

@property
def annotation(self) -> ShapelyPolygon:
"""
Returns the annotation's data representation.

Returns
-------
shapely.geometry.Polygon
The polygon shape.
"""
return self.shape


@dataclass
class Bitmask:
Expand Down Expand Up @@ -222,18 +210,6 @@ def extrema(self) -> tuple[float, float, float, float]:
rows, cols = np.nonzero(self.mask)
return (cols.min(), cols.max(), rows.min(), rows.max())

@property
def annotation(self) -> NDArray[np.bool_]:
"""
Returns the annotation's data representation.

Returns
-------
NDArray[np.bool_]
The binary mask array.
"""
return self.mask


@dataclass
class Detection:
Expand Down
151 changes: 71 additions & 80 deletions lite/valor_lite/object_detection/manager.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,17 +1,10 @@
from collections import defaultdict
from dataclasses import dataclass
from typing import Type

import numpy as np
import valor_lite.object_detection.annotation as annotation
from numpy.typing import NDArray
from tqdm import tqdm
from valor_lite.object_detection.annotation import (
Bitmask,
BoundingBox,
Detection,
Polygon,
)
from valor_lite.object_detection.annotation import Detection
from valor_lite.object_detection.computation import (
compute_bbox_iou,
compute_bitmask_iou,
Expand Down Expand Up @@ -396,74 +389,47 @@ def _add_label(self, label: str) -> int:

return self._evaluator.label_to_index[label]

def _compute_ious_and_cache_pairs(
def _cache_pairs(
self,
uid_index: int,
groundtruths: list,
predictions: list,
annotation_type: Type[BoundingBox] | Type[Polygon] | Type[Bitmask],
ious: NDArray[np.float64],
) -> None:
"""
Compute IOUs between groundtruths and preditions before storing as pairs.

Parameters
----------
uid_index: int
uid_index : int
The index of the detection.
groundtruths: list
groundtruths : list
A list of groundtruths.
predictions: list
predictions : list
A list of predictions.
annotation_type: type[BoundingBox] | type[Polygon] | type[Bitmask]
The type of annotation to compute IOUs for.
ious : NDArray[np.float64]
An array with shape (n_preds, n_gts) containing IOUs.
"""

pairs = list()
n_predictions = len(predictions)
n_groundtruths = len(groundtruths)

all_pairs = np.array(
[
np.array([gann, pann])
for _, _, _, pann in predictions
for _, _, gann in groundtruths
]
)

match annotation_type:
case annotation.BoundingBox:
ious = compute_bbox_iou(all_pairs)
case annotation.Polygon:
ious = compute_polygon_iou(all_pairs)
case annotation.Bitmask:
ious = compute_bitmask_iou(all_pairs)
case _:
raise ValueError(
f"Invalid annotation type `{annotation_type}`."
)

ious = ious.reshape(n_predictions, n_groundtruths)
predictions_with_iou_of_zero = np.where((ious < 1e-9).all(axis=1))[0]
groundtruths_with_iou_of_zero = np.where((ious < 1e-9).all(axis=0))[0]

pairs.extend(
[
np.array(
[
float(uid_index),
float(gidx),
float(pidx),
ious[pidx, gidx],
float(glabel),
float(plabel),
float(score),
]
)
for pidx, plabel, score, _ in predictions
for gidx, glabel, _ in groundtruths
if ious[pidx, gidx] >= 1e-9
]
)
pairs = [
np.array(
[
float(uid_index),
float(gidx),
float(pidx),
ious[pidx, gidx],
float(glabel),
float(plabel),
float(score),
]
)
for pidx, plabel, score in predictions
for gidx, glabel in groundtruths
if ious[pidx, gidx] >= 1e-9
]
pairs.extend(
[
np.array(
Expand Down Expand Up @@ -496,13 +462,12 @@ def _compute_ious_and_cache_pairs(
for index in groundtruths_with_iou_of_zero
]
)

self.pairs.append(np.array(pairs))

def _add_data(
self,
detections: list[Detection],
annotation_type: type[Bitmask] | type[BoundingBox] | type[Polygon],
detection_ious: list[NDArray[np.float64]],
show_progress: bool = False,
):
"""
Expand All @@ -512,13 +477,15 @@ def _add_data(
----------
detections : list[Detection]
A list of Detection objects.
annotation_type : type[Bitmask] | type[BoundingBox] | type[Polygon]
The annotation type to process.
detection_ious : list[NDArray[np.float64]]
A list of arrays containing IOUs per detection.
show_progress : bool, default=False
Toggle for tqdm progress bar.
"""
disable_tqdm = not show_progress
for detection in tqdm(detections, disable=disable_tqdm):
for detection, ious in tqdm(
zip(detections, detection_ious), disable=disable_tqdm
):

# update metadata
self._evaluator.n_datums += 1
Expand All @@ -541,11 +508,6 @@ def _add_data(
predictions = list()

for gidx, gann in enumerate(detection.groundtruths):
if not isinstance(gann, annotation_type):
raise ValueError(
f"Expected {annotation_type}, but annotation is of type {type(gann)}."
)

self._evaluator.groundtruth_examples[uid_index][
gidx
] = gann.extrema
Expand All @@ -556,16 +518,10 @@ def _add_data(
(
gidx,
label_idx,
gann.annotation,
)
)

for pidx, pann in enumerate(detection.predictions):
if not isinstance(pann, annotation_type):
raise ValueError(
f"Expected {annotation_type}, but annotation is of type {type(pann)}."
)

self._evaluator.prediction_examples[uid_index][
pidx
] = pann.extrema
Expand All @@ -577,15 +533,14 @@ def _add_data(
pidx,
label_idx,
pscore,
pann.annotation,
)
)

self._compute_ious_and_cache_pairs(
self._cache_pairs(
uid_index=uid_index,
groundtruths=groundtruths,
predictions=predictions,
annotation_type=annotation_type,
ious=ious,
)

def add_bounding_boxes(
Expand All @@ -603,10 +558,22 @@ def add_bounding_boxes(
show_progress : bool, default=False
Toggle for tqdm progress bar.
"""
ious = [
compute_bbox_iou(
np.array(
[
[gt.extrema, pd.extrema]
for pd in detection.predictions
for gt in detection.groundtruths
]
)
).reshape(len(detection.predictions), len(detection.groundtruths))
for detection in detections
]
return self._add_data(
detections=detections,
detection_ious=ious,
show_progress=show_progress,
annotation_type=BoundingBox,
)

def add_polygons(
Expand All @@ -624,10 +591,22 @@ def add_polygons(
show_progress : bool, default=False
Toggle for tqdm progress bar.
"""
ious = [
compute_polygon_iou(
np.array(
[
[gt.shape, pd.shape] # type: ignore - using the AttributeError as a validator
for pd in detection.predictions
for gt in detection.groundtruths
]
)
).reshape(len(detection.predictions), len(detection.groundtruths))
for detection in detections
]
return self._add_data(
detections=detections,
detection_ious=ious,
show_progress=show_progress,
annotation_type=Polygon,
)

def add_bitmasks(
Expand All @@ -645,10 +624,22 @@ def add_bitmasks(
show_progress : bool, default=False
Toggle for tqdm progress bar.
"""
ious = [
compute_bitmask_iou(
np.array(
[
[gt.mask, pd.mask] # type: ignore - using the AttributeError as a validator
for pd in detection.predictions
for gt in detection.groundtruths
]
)
).reshape(len(detection.predictions), len(detection.groundtruths))
for detection in detections
]
return self._add_data(
detections=detections,
detection_ious=ious,
show_progress=show_progress,
annotation_type=Bitmask,
)

def finalize(self) -> Evaluator:
Expand Down
Loading