crosstyan/OpenGait
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

chore: update demo runtime, tests, and agent docs

Browse Source
This commit is contained in:
crosstyan
2026-03-02 12:33:17 +08:00
parent 1f8f959ad7
commit cbb3284c13
14 changed files with 1491 additions and 236 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/demo/test_pipeline.py
+66 -10
View File
@@ -7,7 +7,7 @@ from pathlib import Path
import subprocess
import sys
import time
from typing import Final, cast
from typing import Final, Literal, cast
from unittest import mock
import numpy as np
@@ -693,9 +693,11 @@ class MockVisualizer:
self,
frame: NDArray[np.uint8],
bbox: tuple[int, int, int, int] | None,
bbox_mask: tuple[int, int, int, int] | None,
track_id: int,
mask_raw: NDArray[np.uint8] | None,
silhouette: NDArray[np.float32] | None,
segmentation_input: NDArray[np.float32] | None,
label: str | None,
confidence: float | None,
fps: float,
@@ -704,9 +706,11 @@ class MockVisualizer:
{
"frame": frame,
"bbox": bbox,
"bbox_mask": bbox_mask,
"track_id": track_id,
"mask_raw": mask_raw,
"silhouette": silhouette,
"segmentation_input": segmentation_input,
"label": label,
"confidence": confidence,
"fps": fps,
@@ -761,9 +765,8 @@ def test_pipeline_visualizer_updates_on_no_detection() -> None:
visualize=True,
)
# Replace the visualizer with our mock
mock_viz = MockVisualizer()
pipeline._visualizer = mock_viz # type: ignore[assignment]
setattr(pipeline, "_visualizer", mock_viz)
# Run pipeline
_ = pipeline.run()
@@ -779,13 +782,14 @@ def test_pipeline_visualizer_updates_on_no_detection() -> None:
for call in mock_viz.update_calls:
assert call["track_id"] == 0 # Default track_id when no detection
assert call["bbox"] is None # No bbox when no detection
assert call["bbox_mask"] is None
assert call["mask_raw"] is None # No mask when no detection
assert call["silhouette"] is None # No silhouette when no detection
assert call["segmentation_input"] is None
assert call["label"] is None # No label when no detection
assert call["confidence"] is None # No confidence when no detection
def test_pipeline_visualizer_uses_cached_detection_on_no_detection() -> None:
"""Test that visualizer reuses last valid detection when current frame has no detection.
@@ -818,8 +822,8 @@ def test_pipeline_visualizer_uses_cached_detection_on_no_detection() -> None:
mock_detector.track.side_effect = [
[mock_result], # Frame 0: valid detection
[mock_result], # Frame 1: valid detection
[], # Frame 2: no detection
[], # Frame 3: no detection
[], # Frame 2: no detection
[], # Frame 3: no detection
]
mock_yolo.return_value = mock_detector
@@ -835,7 +839,12 @@ def test_pipeline_visualizer_uses_cached_detection_on_no_detection() -> None:
dummy_mask = np.random.randint(0, 256, (480, 640), dtype=np.uint8)
dummy_bbox_mask = (100, 100, 200, 300)
dummy_bbox_frame = (100, 100, 200, 300)
mock_select_person.return_value = (dummy_mask, dummy_bbox_mask, dummy_bbox_frame, 1)
mock_select_person.return_value = (
dummy_mask,
dummy_bbox_mask,
dummy_bbox_frame,
1,
)
# Setup mock mask_to_silhouette to return valid silhouette
dummy_silhouette = np.random.rand(64, 44).astype(np.float32)
@@ -856,9 +865,8 @@ def test_pipeline_visualizer_uses_cached_detection_on_no_detection() -> None:
visualize=True,
)
# Replace the visualizer with our mock
mock_viz = MockVisualizer()
pipeline._visualizer = mock_viz # type: ignore[assignment]
setattr(pipeline, "_visualizer", mock_viz)
# Run pipeline
_ = pipeline.run()
@@ -886,9 +894,57 @@ def test_pipeline_visualizer_uses_cached_detection_on_no_detection() -> None:
"not None/blank"
)
# The cached masks should be copies (different objects) to prevent mutation issues
segmentation_inputs = [
call["segmentation_input"] for call in mock_viz.update_calls
]
bbox_mask_calls = [call["bbox_mask"] for call in mock_viz.update_calls]
assert segmentation_inputs[0] is not None
assert segmentation_inputs[1] is not None
assert segmentation_inputs[2] is not None
assert segmentation_inputs[3] is not None
assert bbox_mask_calls[0] == dummy_bbox_mask
assert bbox_mask_calls[1] == dummy_bbox_mask
assert bbox_mask_calls[2] == dummy_bbox_mask
assert bbox_mask_calls[3] == dummy_bbox_mask
if mask_raw_calls[1] is not None and mask_raw_calls[2] is not None:
assert mask_raw_calls[1] is not mask_raw_calls[2], (
"Cached mask should be a copy, not the same object reference"
)
def test_frame_pacer_emission_count_24_to_15() -> None:
from opengait.demo.pipeline import _FramePacer
pacer = _FramePacer(15.0)
interval_ns = int(1_000_000_000 / 24)
emitted = sum(pacer.should_emit(i * interval_ns) for i in range(100))
assert 60 <= emitted <= 65
def test_frame_pacer_requires_positive_target_fps() -> None:
from opengait.demo.pipeline import _FramePacer
with pytest.raises(ValueError, match="target_fps must be positive"):
_FramePacer(0.0)
@pytest.mark.parametrize(
("window", "stride", "mode", "expected"),
[
(30, 30, "manual", 30),
(30, 7, "manual", 7),
(30, 30, "sliding", 1),
(30, 1, "chunked", 30),
(15, 3, "chunked", 15),
],
)
def test_resolve_stride_modes(
window: int,
stride: int,
mode: Literal["manual", "sliding", "chunked"],
expected: int,
) -> None:
from opengait.demo.pipeline import resolve_stride
assert resolve_stride(window, stride, mode) == expected
tests/demo/test_visualizer.py
+171
View File
@@ -0,0 +1,171 @@
from __future__ import annotations
from pathlib import Path
from typing import cast
from unittest import mock
import numpy as np
import pytest
from opengait.demo.input import create_source
from opengait.demo.visualizer import (
DISPLAY_HEIGHT,
DISPLAY_WIDTH,
ImageArray,
OpenCVVisualizer,
)
from opengait.demo.window import select_person
REPO_ROOT = Path(__file__).resolve().parents[2]
SAMPLE_VIDEO_PATH = REPO_ROOT / "assets" / "sample.mp4"
YOLO_MODEL_PATH = REPO_ROOT / "ckpt" / "yolo11n-seg.pt"
def test_prepare_raw_view_float_mask_has_visible_signal() -> None:
viz = OpenCVVisualizer()
mask_float = np.zeros((64, 64), dtype=np.float32)
mask_float[16:48, 16:48] = 1.0
rendered = viz._prepare_raw_view(cast(ImageArray, mask_float))
assert rendered.dtype == np.uint8
assert rendered.shape == (256, 176, 3)
mask_zero = np.zeros((64, 64), dtype=np.float32)
rendered_zero = viz._prepare_raw_view(cast(ImageArray, mask_zero))
roi = slice(0, DISPLAY_HEIGHT - 40)
diff = np.abs(rendered[roi].astype(np.int16) - rendered_zero[roi].astype(np.int16))
assert int(np.count_nonzero(diff)) > 0
def test_prepare_raw_view_handles_values_slightly_above_one() -> None:
viz = OpenCVVisualizer()
mask = np.zeros((64, 64), dtype=np.float32)
mask[20:40, 20:40] = 1.0001
rendered = viz._prepare_raw_view(cast(ImageArray, mask))
roi = rendered[: DISPLAY_HEIGHT - 40, :, 0]
assert int(np.count_nonzero(roi)) > 0
def test_segmentation_view_is_normalized_only_shape() -> None:
viz = OpenCVVisualizer()
mask = np.zeros((480, 640), dtype=np.uint8)
sil = np.random.rand(64, 44).astype(np.float32)
seg = viz._prepare_segmentation_view(cast(ImageArray, mask), sil, (0, 0, 100, 100))
assert seg.shape == (DISPLAY_HEIGHT, DISPLAY_WIDTH, 3)
def test_update_toggles_raw_window_with_r_key() -> None:
viz = OpenCVVisualizer()
frame = np.zeros((240, 320, 3), dtype=np.uint8)
mask = np.zeros((240, 320), dtype=np.uint8)
mask[20:100, 30:120] = 255
sil = np.random.rand(64, 44).astype(np.float32)
seg_input = np.random.rand(4, 64, 44).astype(np.float32)
with (
mock.patch("cv2.namedWindow") as named_window,
mock.patch("cv2.imshow"),
mock.patch("cv2.destroyWindow") as destroy_window,
mock.patch("cv2.waitKey", side_effect=[ord("r"), ord("r"), ord("q")]),
):
assert viz.update(
frame,
(10, 10, 120, 150),
(10, 10, 120, 150),
1,
cast(ImageArray, mask),
sil,
seg_input,
None,
None,
15.0,
)
assert viz.show_raw_window is True
assert viz._raw_window_created is True
assert viz.update(
frame,
(10, 10, 120, 150),
(10, 10, 120, 150),
1,
cast(ImageArray, mask),
sil,
seg_input,
None,
None,
15.0,
)
assert viz.show_raw_window is False
assert viz._raw_window_created is False
assert destroy_window.called
assert (
viz.update(
frame,
(10, 10, 120, 150),
(10, 10, 120, 150),
1,
cast(ImageArray, mask),
sil,
seg_input,
None,
None,
15.0,
)
is False
)
assert named_window.called
def test_sample_video_raw_mask_shape_range_and_render_signal() -> None:
if not SAMPLE_VIDEO_PATH.is_file():
pytest.skip(f"Missing sample video: {SAMPLE_VIDEO_PATH}")
if not YOLO_MODEL_PATH.is_file():
pytest.skip(f"Missing YOLO model file: {YOLO_MODEL_PATH}")
ultralytics = pytest.importorskip("ultralytics")
yolo_cls = getattr(ultralytics, "YOLO")
viz = OpenCVVisualizer()
detector = yolo_cls(str(YOLO_MODEL_PATH))
masks_seen = 0
rendered_nonzero: list[int] = []
for frame, _meta in create_source(str(SAMPLE_VIDEO_PATH), max_frames=30):
detections = detector.track(
frame,
persist=True,
verbose=False,
classes=[0],
device="cpu",
)
if not isinstance(detections, list) or not detections:
continue
selected = select_person(detections[0])
if selected is None:
continue
mask_raw, _, _, _ = selected
masks_seen += 1
arr = np.asarray(mask_raw)
assert arr.ndim == 2
assert arr.shape[0] > 0 and arr.shape[1] > 0
assert np.issubdtype(arr.dtype, np.number)
assert float(arr.min()) >= 0.0
assert float(arr.max()) <= 255.0
assert int(np.count_nonzero(arr)) > 0
rendered = viz._prepare_raw_view(arr)
roi = rendered[: DISPLAY_HEIGHT - 40, :, 0]
rendered_nonzero.append(int(np.count_nonzero(roi)))
assert masks_seen > 0
assert min(rendered_nonzero) > 0