fix: is_box_in_image function

Signed-off-by: ktro2828 <[email protected]>

t4_devkit/common/geometry.py

@@ -68,7 +68,7 @@ def view_points(
         points = points[:3, :]
 
     if normalize:
-        points = points / points[2:3, :].repeat(3, 0).reshape(3, nbr_points)
+        points /= points[2:3, :]
 
     return points
 
@@ -90,21 +90,21 @@ def is_box_in_image(
     Returns:
         Return True if visibility condition is satisfied.
     """
-    corners_3d = box.corners()
-    corners_on_img = view_points(corners_3d.T, intrinsic, normalize=True)[:2, :]
+    corners_3d = box.corners().T  # (3, 8)
+    corners_on_img = view_points(corners_3d, intrinsic, normalize=True)[:2, :]
 
     img_w, img_h = img_size
-    is_visible = np.logical_and(corners_on_img[0, :] > 0, corners_on_img[0, :] < img_w)
-    is_visible = np.logical_and(is_visible, corners_on_img[1, :] < img_h)
-    is_visible = np.logical_and(is_visible, corners_on_img[1, :] > 0)
-    is_visible = np.logical_and(is_visible, corners_on_img[2, :] > 1)
+    is_visible = np.logical_and(corners_on_img[0, :] >= 0, corners_on_img[0, :] <= img_w)
+    is_visible = np.logical_and(is_visible, corners_on_img[1, :] <= img_h)
+    is_visible = np.logical_and(is_visible, corners_on_img[1, :] >= 0)
+    is_visible = np.logical_and(is_visible, corners_3d[2, :] > 1)
 
     in_front = corners_3d[2, :] > 0.1  # True if a corner is at least 0.1 meter in front of camera.
 
     if visibility == VisibilityLevel.FULL:
-        return all(is_visible) and all(in_front)
+        return np.all(is_visible) and np.all(in_front)
     elif visibility in (VisibilityLevel.MOST, VisibilityLevel.PARTIAL):
-        return any(is_visible)
+        return np.any(is_visible)
     elif visibility == VisibilityLevel.NONE:
         return True
     else:

tests/common/test_geometry.py

@@ -1,6 +1,7 @@
 import numpy as np
 
-from t4_devkit.common.geometry import view_points
+from t4_devkit.common.geometry import is_box_in_image, view_points
+from t4_devkit.schema import VisibilityLevel
 
 
 def test_view_points_by_perspective_projection() -> None:
@@ -77,3 +78,46 @@ def test_view_points_with_distortion() -> None:
     )
 
     assert np.allclose(project, expect)
+
+
+# TODO(ktro2828): add unit testing for VisibilityLevel.FULL
+
+
+def test_box_partial_visible(dummy_box3d, dummy_camera_calibration) -> None:
+    """Test `is_box_in_image` function in the case of the box is partially visible."""
+    img_size, intrinsic = dummy_camera_calibration
+
+    dummy_box3d.position = (0.0, 0.0, 2.0)
+    assert is_box_in_image(
+        dummy_box3d,
+        intrinsic=intrinsic,
+        img_size=img_size,
+        visibility=VisibilityLevel.PARTIAL,
+    )
+
+
+def test_box_not_visible(dummy_box3d, dummy_camera_calibration) -> None:
+    """Test `is_box_in_image` function in the case of the box is not visible."""
+    img_size, intrinsic = dummy_camera_calibration
+
+    dummy_box3d.position = (100.0, 100.0, 1.0)
+    assert not is_box_in_image(
+        dummy_box3d,
+        intrinsic=intrinsic,
+        img_size=img_size,
+        visibility=VisibilityLevel.PARTIAL,
+    )
+
+
+def test_box_behind_camera(dummy_box3d, dummy_camera_calibration) -> None:
+    """Test `is_box_in_image` function in the case of the box is behind of the camera."""
+
+    img_size, intrinsic = dummy_camera_calibration
+
+    dummy_box3d.position = (100.0, 100.0, -1.0)
+    assert not is_box_in_image(
+        dummy_box3d,
+        intrinsic=intrinsic,
+        img_size=img_size,
+        visibility=VisibilityLevel.FULL,
+    )

tests/conftest.py

@@ -1,3 +1,8 @@
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
+import numpy as np
 import pytest
 from pyquaternion import Quaternion
 
@@ -12,6 +17,9 @@
     TransformBuffer,
 )
 
+if TYPE_CHECKING:
+    from t4_devkit.typing import NDArrayFloat
+
 
 @pytest.fixture(scope="module")
 def label2id() -> dict[str, int]:
@@ -176,3 +184,18 @@ def dummy_tf_buffer() -> TransformBuffer:
     )
 
     return tf_buffer
+
+
+@pytest.fixture(scope="function")
+def dummy_camera_calibration() -> tuple[tuple[int, int], NDArrayFloat]:
+    img_size = (1280, 720)
+
+    intrinsic = np.array(
+        [
+            [1000, 0, 640],
+            [0, 1000, 360],
+            [0, 0, 1],
+        ]
+    )
+
+    return img_size, intrinsic