def test_one_person_two_stage_pipe_high_face_confidence():
"""Detect a person in 1st stage and a face in 2nd stage."""
object_config = _object_detect_config()
face_config = _face_detect_config()
result = None
def sample_callback(image=None, inference_result=None, **kwargs):
nonlocal result
result = inference_result
object_detector = ObjectDetector(**object_config)
face_detector = FaceDetector(**face_config)
object_detector.connect_to_next_element(face_detector)
output = _OutPipeElement(sample_callback=sample_callback)
face_detector.connect_to_next_element(output)
img = _get_image(file_name='person-face.jpg')
object_detector.receive_next_sample(image=img)
assert result
assert len(result) == 1
label = result[0]['label']
confidence = result[0]['confidence']
(x0, y0) = result[0]['box']['xmin'], result[0]['box']['ymin']
(x1, y1) = result[0]['box']['xmax'], result[0]['box']['ymax']
assert label == 'person'
assert confidence > 0.9
assert x0 > 0 and x0 < x1
assert y0 > 0 and y0 < y1
def test_thermal_one_person_miss_face_two_stage_pipe():
"""Expect to detect a person but not a face."""
object_config = _object_detect_config()
face_config = _face_detect_config()
result = None
def sample_callback(image=None, inference_result=None, **kwargs):
nonlocal result
result = inference_result
# test stage one, obect detection -> out
object_detector = ObjectDetector(**object_config)
output = _OutPipeElement(sample_callback=sample_callback)
object_detector.connect_to_next_element(output)
img = _get_image(file_name="person_thermal_bw.jpg")
object_detector.receive_next_sample(image=img)
assert result
assert len(result) == 1
label = result[0]["label"]
confidence = result[0]["confidence"]
(x0, y0) = result[0]["box"]["xmin"], result[0]["box"]["ymin"]
(x1, y1) = result[0]["box"]["xmax"], result[0]["box"]["ymax"]
assert label == "person"
assert confidence > 0.8
assert x0 > 0 and x0 < x1
assert y0 > 0 and y0 < y1
# test stage 2, rearrange pipe elements: object->face->out
face_detector = FaceDetector(**face_config)
object_detector.connect_to_next_element(face_detector)
face_detector.connect_to_next_element(output)
object_detector.receive_next_sample(image=img)
assert not result
def test2_one_person_high_confidence_face_low_confidence_two_stage_pipe():
"""Expect to detect a person but not a face."""
object_config = _object_detect_config()
face_config = _face_detect_config()
result = None
def sample_callback(image=None, inference_result=None, **kwargs):
nonlocal result
result = inference_result
# test stage one, obect detection -> out
object_detector = ObjectDetector(**object_config)
output = _OutPipeElement(sample_callback=sample_callback)
object_detector.connect_to_next_element(output)
img = _get_image(file_name='person-face2.jpg')
object_detector.receive_next_sample(image=img)
assert result
assert len(result) == 1
label = result[0]['label']
confidence = result[0]['confidence']
(x0, y0) = result[0]['box']['xmin'], result[0]['box']['ymin']
(x1, y1) = result[0]['box']['xmax'], result[0]['box']['ymax']
assert label == 'person'
assert confidence > 0.9
assert x0 > 0 and x0 < x1
assert y0 > 0 and y0 < y1
# test stage 2, rearrange pipe elements: object->face->out
face_detector = FaceDetector(**face_config)
object_detector.connect_to_next_element(face_detector)
face_detector.connect_to_next_element(output)
object_detector.receive_next_sample(image=img)
assert not result
def test_one_person_high_confidence_face_low_confidence_two_stage_pipe():
"""Expect to detect a person but not a face."""
object_config = _object_detect_config()
face_config = _face_detect_config()
result = None
def sample_callback(image=None, inference_result=None):
nonlocal result
result = inference_result
# test stage one, obect detection -> out
object_detector = ObjectDetector(element_config=object_config)
output = _OutPipeElement(sample_callback=sample_callback)
object_detector.connect_to_next_element(output)
img = _get_image(file_name='person.jpg')
object_detector.receive_next_sample(image=img)
assert result
assert len(result) == 1
category, confidence, (x0, y0, x1, y1) = result[0]
assert category == 'person'
assert confidence > 0.9
assert x0 > 0 and x0 < x1
assert y0 > 0 and y0 < y1
# test stage 2, rearrange pipe elements: object->face->out
face_detector = FaceDetector(element_config=face_config)
object_detector.connect_to_next_element(face_detector)
face_detector.connect_to_next_element(output)
object_detector.receive_next_sample(image=img)
assert not result
def test_bad_sample_good_sample():
"""One bad sample should not prevent good samples from being processed."""
config = _face_detect_config()
result = 'Something'
def sample_callback(image=None, inference_result=None, **kwargs):
nonlocal result
result = inference_result
face_detector = FaceDetector(**config)
output = _OutPipeElement(sample_callback=sample_callback)
face_detector.connect_to_next_element(output)
# bad sample
face_detector.receive_next_sample(image=None,
inference_result=[
('person', 1, [-1, -2, -3, -4]),
])
# good sample
img = _get_image(file_name='person-face.jpg')
face_detector.receive_next_sample(image=img,
inference_result=[
('person', 1, [0, 0, 1, 1]),
])
assert result
assert len(result) == 1
label, confidence, (x0, y0, x1, y1) = result[0]
assert label == 'person'
assert confidence > 0.8
assert x0 > 0 and x0 < x1
assert y0 > 0 and y0 < y1
def test_one_person_no_face_two_stage():
"""Expect to detect one person."""
object_config = _object_detect_config()
face_config = _face_detect_config()
result = None
def sample_callback(image=None, inference_result=None, **kwargs):
nonlocal result
result = inference_result
object_detector = ObjectDetector(**object_config)
face_detector = FaceDetector(**face_config)
object_detector.connect_to_next_element(face_detector)
output = _OutPipeElement(sample_callback=sample_callback)
face_detector.connect_to_next_element(output)
img = _get_image(file_name='person-no-face.jpg')
object_detector.receive_next_sample(image=img)
assert not result
def test_one_person_two_stage_pipe_low_person_confidence():
"""Fail to detect person in 1st stage hence no face in 2nd stage."""
object_config = _object_detect_config()
face_config = _face_detect_config()
result = None
def sample_callback(image=None, inference_result=None):
nonlocal result
result = inference_result
object_detector = ObjectDetector(element_config=object_config)
face_detector = FaceDetector(element_config=face_config)
object_detector.connect_to_next_element(face_detector)
output = _OutPipeElement(sample_callback=sample_callback)
face_detector.connect_to_next_element(output)
img = _get_image(file_name='person-face.jpg')
object_detector.receive_next_sample(image=img)
assert not result
def test_two_person_high_confidence_one_face_high_confidence_two_stage_pipe():
"""Expect to detect two persons but only one face."""
object_config = _object_detect_config()
face_config = _face_detect_config()
result = None
def sample_callback(image=None, inference_result=None, **kwargs):
nonlocal result
result = inference_result
# test stage one, obect detection -> out
object_detector = ObjectDetector(**object_config)
output = _OutPipeElement(sample_callback=sample_callback)
object_detector.connect_to_next_element(output)
img = _get_image(file_name='person2-face1.jpg')
object_detector.receive_next_sample(image=img)
assert result
assert len(result) == 2
label, confidence, (x0, y0, x1, y1) = result[0]
assert label == 'person'
assert confidence > 0.9
assert x0 > 0 and x0 < x1
assert y0 > 0 and y0 < y1
label, confidence, (x0, y0, x1, y1) = result[1]
assert label == 'person'
assert confidence > 0.9
assert x0 > 0 and x0 < x1
assert y0 > 0 and y0 < y1
# test stage 2, rearrange pipe elements: object->face->out
face_detector = FaceDetector(**face_config)
object_detector.connect_to_next_element(face_detector)
face_detector.connect_to_next_element(output)
object_detector.receive_next_sample(image=img)
assert result
assert len(result) == 1
label, confidence, (x0, y0, x1, y1) = result[0]
assert label == 'person'
assert confidence > 0.6
assert x0 > 0 and x0 < x1
assert y0 > 0 and y0 < y1
def test_one_person_face_high_confidence_one_stage_pipe():
"""Expect to detect one person face."""
face_config = _face_detect_config()
result = None
def sample_callback(image=None, inference_result=None, **kwargs):
nonlocal result
result = inference_result
face_detector = FaceDetector(**face_config)
output = _OutPipeElement(sample_callback=sample_callback)
face_detector.connect_to_next_element(output)
img = _get_image(file_name='person-face.jpg')
face_detector.receive_next_sample(image=img,
inference_result=[{
'label': 'person',
'confidence': 1,
'box': {
'xmin': 0,
'ymin': 0,
'xmax': 1,
'ymax': 1
}
}])
assert result
assert len(result) == 1
label = result[0]['label']
confidence = result[0]['confidence']
(x0, y0) = result[0]['box']['xmin'], result[0]['box']['ymin']
(x1, y1) = result[0]['box']['xmax'], result[0]['box']['ymax']
assert label == 'person'
assert confidence > 0.8
assert x0 > 0 and x0 < x1
assert y0 > 0 and y0 < y1
def test_two_person_with_faces_no_confidence_one_stage_pipe():
"""No faces detected when the whole jpg is fed directly to face detect.
Illustrates the point of piping models from more generic to specialized.
The object detection model detects people and feeds a zoomed in image
to the face detector. Instead of resizing 1280x720 to 320x320, which
results in face detector missing faces, the piping feeds a smaller region
of the original image where a person was detected to face detection.
That resizes approximately a 500x400 image to 320x320, which preserves
more feature information.
"""
face_config = _face_detect_config()
result = None
def sample_callback(image=None, inference_result=None, **kwargs):
nonlocal result
result = inference_result
face_detector = FaceDetector(**face_config)
output = _OutPipeElement(sample_callback=sample_callback)
face_detector.connect_to_next_element(output)
img = _get_image(file_name='person2-face1.jpg')
face_detector.receive_next_sample(image=img,
inference_result=[
('person', 1, [0, 0, 1, 1]),
])
assert not result
def test_one_person_face_high_confidence_one_stage_pipe():
"""Expect to detect one person face."""
face_config = _face_detect_config()
result = None
def sample_callback(image=None, inference_result=None, **kwargs):
nonlocal result
result = inference_result
face_detector = FaceDetector(**face_config)
output = _OutPipeElement(sample_callback=sample_callback)
face_detector.connect_to_next_element(output)
img = _get_image(file_name='person-face.jpg')
face_detector.receive_next_sample(image=img,
inference_result=[
('person', 1, [0, 0, 1, 1]),
])
assert result
assert len(result) == 1
label, confidence, (x0, y0, x1, y1) = result[0]
assert label == 'person'
assert confidence > 0.8
assert x0 > 0 and x0 < x1
assert y0 > 0 and y0 < y1
def test_one_person_face_high_confidence_one_stage_pipe():
"""Expect to detect one person face."""
face_config = _face_detect_config()
result = None
def sample_callback(image=None, inference_result=None, **kwargs):
nonlocal result
result = inference_result
face_detector = FaceDetector(**face_config)
output = _OutPipeElement(sample_callback=sample_callback)
face_detector.connect_to_next_element(output)
img = _get_image(file_name="person-face.jpg")
face_detector.receive_next_sample(
image=img,
inference_result=[
{
"label": "person",
"confidence": 1,
"box": {"xmin": 0, "ymin": 0, "xmax": 1, "ymax": 1},
}
],
)
assert result
assert len(result) == 1
label = result[0]["label"]
confidence = result[0]["confidence"]
(x0, y0) = result[0]["box"]["xmin"], result[0]["box"]["ymin"]
(x1, y1) = result[0]["box"]["xmax"], result[0]["box"]["ymax"]
assert label == "person"
assert confidence > 0.8
assert x0 > 0 and x0 < x1
assert y0 > 0 and y0 < y1
def test_model_inputs():
"""Verify against known model inputs."""
config = _face_detect_config()
face_detector = FaceDetector(**config)
tfe = face_detector._tfengine
samples = tfe.input_details[0]['shape'][0]
assert samples == 1
height = tfe.input_details[0]['shape'][1]
assert height == 320
width = tfe.input_details[0]['shape'][2]
assert width == 320
colors = tfe.input_details[0]['shape'][3]
assert colors == 3
def test_bad_sample_good_sample():
"""One bad sample should not prevent good samples from being processed."""
config = _face_detect_config()
result = 'Something'
def sample_callback(image=None, inference_result=None, **kwargs):
nonlocal result
result = inference_result
face_detector = FaceDetector(**config)
output = _OutPipeElement(sample_callback=sample_callback)
face_detector.connect_to_next_element(output)
# bad sample
face_detector.receive_next_sample(image=None,
inference_result=[{
'label': 'person',
'confidence': 1,
'box': {
'xmin': -1,
'ymin': -2,
'xmax': -3,
'ymax': -4
}
}])
# good sample
img = _get_image(file_name='person-face.jpg')
face_detector.receive_next_sample(image=img,
inference_result=[{
'label': 'person',
'confidence': 1,
'box': {
'xmin': 0,
'ymin': 0,
'xmax': 1,
'ymax': 1
}
}])
assert result
assert len(result) == 1
label = result[0]['label']
confidence = result[0]['confidence']
(x0, y0) = result[0]['box']['xmin'], result[0]['box']['ymin']
(x1, y1) = result[0]['box']['xmax'], result[0]['box']['ymax']
assert label == 'person'
assert confidence > 0.8
assert x0 > 0 and x0 < x1
assert y0 > 0 and y0 < y1
def test_bad_sample_good_sample():
"""One bad sample should not prevent good samples from being processed."""
config = _face_detect_config()
result = "Something"
def sample_callback(image=None, inference_result=None, **kwargs):
nonlocal result
result = inference_result
face_detector = FaceDetector(**config)
output = _OutPipeElement(sample_callback=sample_callback)
face_detector.connect_to_next_element(output)
# bad sample
face_detector.receive_next_sample(
image=None,
inference_result=[
{
"label": "person",
"confidence": 1,
"box": {"xmin": -1, "ymin": -2, "xmax": -3, "ymax": -4},
}
],
)
# good sample
img = _get_image(file_name="person-face.jpg")
face_detector.receive_next_sample(
image=img,
inference_result=[
{
"label": "person",
"confidence": 1,
"box": {"xmin": 0, "ymin": 0, "xmax": 1, "ymax": 1},
}
],
)
assert result
assert len(result) == 1
label = result[0]["label"]
confidence = result[0]["confidence"]
(x0, y0) = result[0]["box"]["xmin"], result[0]["box"]["ymin"]
(x1, y1) = result[0]["box"]["xmax"], result[0]["box"]["ymax"]
assert label == "person"
assert confidence > 0.8
assert x0 > 0 and x0 < x1
assert y0 > 0 and y0 < y1
def test_model_outputs():
"""Verify against known model outputs."""
config = _face_detect_config()
face_detector = FaceDetector(**config)
tfe = face_detector._tfengine
assert tfe.output_details[0]['shape'][0] == 1
scores = tfe.output_details[0]['shape'][1]
assert scores == 50
assert tfe.output_details[1]['shape'][0] == 1
boxes = tfe.output_details[1]['shape'][1]
assert boxes == 50
assert tfe.output_details[2]['shape'][0] == 1
labels = tfe.output_details[2]['shape'][1]
assert labels == 50
num = tfe.output_details[3]['shape'][0]
assert num == 1
def test_no_sample():
"""Expect element to pass empty sample to next element."""
config = _object_detect_config()
result = 'Something'
def sample_callback(image=None, inference_result=None, **kwargs):
nonlocal result
result = image is None and inference_result is None
face_detector = FaceDetector(**config)
output = _OutPipeElement(sample_callback=sample_callback)
face_detector.connect_to_next_element(output)
face_detector.receive_next_sample()
assert result is True
def test_background_image_no_person():
"""Expect to not detect anything interesting in a background image."""
config = _face_detect_config()
result = None
def sample_callback(image=None, inference_result=None, **kwargs):
nonlocal result
result = not image and not inference_result
face_detector = FaceDetector(**config)
output = _OutPipeElement(sample_callback=sample_callback)
face_detector.connect_to_next_element(output)
img = _get_image(file_name='background.jpg')
face_detector.receive_next_sample(image=img)
assert result is True
