def predict(self,
image_path,
threshold=0.1,
nms_overlap=0.1,
return_scores=True,
visualize=False,
resize=False):
"""
Predicts and displays the results of a trained model on a single image.
===================== ===========================================
**Argument** **Description**
--------------------- -------------------------------------------
image_path Required. Path to the image file to make the
predictions on.
--------------------- -------------------------------------------
thresh Optional float. The probabilty above which
a detection will be considered valid.
--------------------- -------------------------------------------
nms_overlap Optional float. The intersection over union
threshold with other predicted bounding
boxes, above which the box with the highest
score will be considered a true positive.
--------------------- -------------------------------------------
return_scores Optional boolean.
Will return the probability scores of the
bounding box predictions if True.
--------------------- -------------------------------------------
visualize Optional boolean. Displays the image with
predicted bounding boxes if True.
--------------------- -------------------------------------------
resize Optional boolean. Resizes the image to the same size
(chip_size parameter in prepare_data) that the model was trained on,
before detecting objects.
Note that if resize_to parameter was used in prepare_data,
the image is resized to that size instead.
By default, this parameter is false and the detections are run
in a sliding window fashion by applying the model on cropped sections
of the image (of the same size as the model was trained on).
===================== ===========================================
:returns: 'List' of xmin, ymin, width, height of predicted bounding boxes on the given image
"""
if not HAS_OPENCV:
raise Exception(
"This function requires opencv 4.0.1.24. Install it using pip install opencv-python==4.0.1.24"
)
if not HAS_PIL:
raise Exception(
"This function requires PIL. Please install it via pip or conda"
)
if isinstance(image_path, str):
image = cv2.imread(image_path)
else:
image = image_path
orig_height, orig_width, _ = image.shape
orig_frame = image.copy()
if resize and self._data.resize_to is None\
and self._data.chip_size is not None:
image = cv2.resize(image,
(self._data.chip_size, self._data.chip_size))
if self._data.resize_to is not None:
if isinstance(self._data.resize_to, tuple):
image = cv2.resize(image, self._data.resize_to)
else:
image = cv2.resize(
image, (self._data.resize_to, self._data.resize_to))
height, width, _ = image.shape
if self._data.chip_size is not None:
chips = _get_image_chips(image, self._data.chip_size)
else:
chips = [{
'width': width,
'height': height,
'xmin': 0,
'ymin': 0,
'chip': image,
'predictions': []
}]
valid_tfms = self._data.valid_ds.tfms
self._data.valid_ds.tfms = []
include_pad_detections = False
if len(chips) == 1:
include_pad_detections = True
for chip in chips:
frame = Image(
pil2tensor(PIL.Image.fromarray(
cv2.cvtColor(chip['chip'], cv2.COLOR_BGR2RGB)),
dtype=np.float32).div_(255))
self.learn.predicting = True
bbox = self.learn.predict(frame,
thresh=threshold,
nms_overlap=nms_overlap,
ret_scores=True,
model=self)[0]
if bbox:
scores = bbox.scores
bboxes, lbls = bbox._compute_boxes()
bboxes.add_(1).mul_(
torch.tensor([
chip['height'] / 2, chip['width'] / 2,
chip['height'] / 2, chip['width'] / 2
])).long()
for index, bbox in enumerate(bboxes):
if lbls is not None:
label = lbls[index]
else:
label = 'Default'
data = bb2hw(bbox)
if include_pad_detections or not _exclude_detection(
(data[0], data[1], data[2], data[3]), chip['width'],
chip['height']):
chip['predictions'].append({
'xmin':
data[0],
'ymin':
data[1],
'width':
data[2],
'height':
data[3],
'score':
float(scores[index]),
'label':
label
})
self._data.valid_ds.tfms = valid_tfms
predictions, labels, scores = _get_transformed_predictions(chips)
# Scale the predictions to original image and clip the predictions to image dims
y_ratio = orig_height / height
x_ratio = orig_width / width
for index, prediction in enumerate(predictions):
prediction[0] = prediction[0] * x_ratio
prediction[1] = prediction[1] * y_ratio
prediction[2] = prediction[2] * x_ratio
prediction[3] = prediction[3] * y_ratio
# Clip xmin
if prediction[0] < 0:
prediction[2] = prediction[2] + prediction[0]
prediction[0] = 1
# Clip width when xmax greater than original width
if prediction[0] + prediction[2] > orig_width:
prediction[2] = (prediction[0] + prediction[2]) - orig_width
# Clip ymin
if prediction[1] < 0:
prediction[3] = prediction[3] + prediction[1]
prediction[1] = 1
# Clip height when ymax greater than original height
if prediction[1] + prediction[3] > orig_height:
prediction[3] = (prediction[1] + prediction[3]) - orig_height
predictions[index] = [
prediction[0], prediction[1], prediction[2], prediction[3]
]
if visualize:
image = _draw_predictions(orig_frame,
predictions,
labels,
color=(255, 0, 0),
fontface=2,
thickness=1)
import matplotlib.pyplot as plt
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
if getattr(self._data, "_is_coco", "") == True:
figsize = (20, 20)
else:
figsize = (4, 4)
fig, ax = plt.subplots(1, 1, figsize=figsize)
ax.imshow(image)
if return_scores:
return predictions, labels, scores
else:
return predictions, labels
def predict(
self,
image_path,
threshold=0.5,
nms_overlap=0.1,
return_scores=False,
visualize=False
):
"""
Runs prediction on a video and appends the output VMTI predictions in the metadata file.
===================== ===========================================
**Argument** **Description**
--------------------- -------------------------------------------
image_path Required. Path to the image file to make the
predictions on.
--------------------- -------------------------------------------
threshold Optional float. The probability above which
a detection will be considered valid.
--------------------- -------------------------------------------
nms_overlap Optional float. The intersection over union
threshold with other predicted bounding
boxes, above which the box with the highest
score will be considered a true positive.
--------------------- -------------------------------------------
return_scores Optional boolean. Will return the probability
scores of the bounding box predictions if True.
--------------------- -------------------------------------------
visualize Optional boolean. Displays the image with
predicted bounding boxes if True.
===================== ===========================================
:returns: 'List' of xmin, ymin, width, height of predicted bounding boxes on the given image
"""
if not HAS_OPENCV:
raise Exception("This function requires opencv 4.0.1.24. Install it using pip install opencv-python==4.0.1.24")
if isinstance(image_path, str):
image = cv2.imread(image_path)
else:
image = image_path
orig_height, orig_width, _ = image.shape
orig_frame = image.copy()
if self._data.resize_to is not None:
if isinstance(self._data.resize_to, tuple):
image = cv2.resize(image, self._data.resize_to)
else:
image = cv2.resize(image, (self._data.resize_to, self._data.resize_to))
height, width, _ = image.shape
if self._data.chip_size is not None:
chips = _get_image_chips(image, self._data.chip_size)
else:
chips = [{'width': width, 'height': height, 'xmin': 0, 'ymin': 0, 'chip': image, 'predictions': []}]
valid_tfms = self._data.valid_ds.tfms
self._data.valid_ds.tfms = []
for chip in chips:
frame = Image(pil2tensor(PIL.Image.fromarray(cv2.cvtColor(chip['chip'], cv2.COLOR_BGR2RGB)), dtype=np.float32).div_(255))
bbox = self.learn.predict(frame, thresh=threshold, nms_overlap=nms_overlap, ret_scores=True, ssd=self)[0]
if bbox:
scores = bbox.scores
bboxes, lbls = bbox._compute_boxes()
bboxes.add_(1).mul_(torch.tensor([chip['height'] / 2, chip['width'] / 2, chip['height'] / 2, chip['width'] / 2])).long()
for index, bbox in enumerate(bboxes):
if lbls is not None:
label = lbls[index]
else:
label = 'Default'
data = bb2hw(bbox)
if not _exclude_detection((data[0], data[1], data[2], data[3]), chip['width'], chip['height']):
chip['predictions'].append({
'xmin': data[0],
'ymin': data[1],
'width': data[2],
'height': data[3],
'score': float(scores[index]),
'label': label
})
self._data.valid_ds.tfms = valid_tfms
predictions, labels, scores = _get_transformed_predictions(chips)
y_ratio = orig_height/height
x_ratio = orig_width/width
for index, prediction in enumerate(predictions):
prediction[0] = prediction[0]*x_ratio
prediction[1] = prediction[1]*y_ratio
prediction[2] = prediction[2]*x_ratio
prediction[3] = prediction[3]*y_ratio
# Clip xmin
if prediction[0] < 0:
prediction[2] = prediction[2] + prediction[0]
prediction[0] = 1
# Clip width when xmax greater than original width
if prediction[0] + prediction[2] > orig_width:
prediction[2] = (prediction[0] + prediction[2]) - orig_width
# Clip ymin
if prediction[1] < 0:
prediction[3] = prediction[3] + prediction[1]
prediction[1] = 1
# Clip height when ymax greater than original height
if prediction[1] + prediction[3] > orig_height:
prediction[3] = (prediction[1] + prediction[3]) - orig_height
predictions[index] = [
prediction[0],
prediction[1],
prediction[2],
prediction[3]
]
if visualize:
image = _draw_predictions(orig_frame, predictions, labels)
import matplotlib.pyplot as plt
plt.xticks([])
plt.yticks([])
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
plt.imshow(PIL.Image.fromarray(image))
if return_scores:
return predictions, labels, scores
else:
return predictions, labels