def compute_gt_annotations(anchors,
annotations,
negative_overlap=0.4,
positive_overlap=0.5):
""" Obtain indices of gt annotations with the greatest overlap.
Args
anchors: np.array of annotations of shape (N, 4) for (x1, y1, x2, y2).
annotations: np.array of shape (N, 5) for (x1, y1, x2, y2, label).
negative_overlap: IoU overlap for negative anchors (all anchors with overlap < negative_overlap are negative).
positive_overlap: IoU overlap or positive anchors (all anchors with overlap > positive_overlap are positive).
Returns
positive_indices: indices of positive anchors
ignore_indices: indices of ignored anchors
argmax_overlaps_inds: ordered overlaps indices
"""
overlaps = compute_overlap(anchors.astype(np.float64),
annotations.astype(np.float64))
argmax_overlaps_inds = np.argmax(overlaps, axis=1)
max_overlaps = overlaps[np.arange(overlaps.shape[0]), argmax_overlaps_inds]
# assign "dont care" labels
positive_indices = max_overlaps >= positive_overlap
ignore_indices = (max_overlaps > negative_overlap) & ~positive_indices
return positive_indices, ignore_indices, argmax_overlaps_inds
def find_overlap_artifacts(artifacts):
'''Find indices of bad_epochs that overlap with each other.'''
overlap_indices = []
for i, a in enumerate(artifacts):
if i > len(artifacts) - 2: pass
elif utils.compute_overlap(a.st_sample, a.et_sample,
artifacts[i + 1].st_sample,
artifacts[i + 1].et_sample) > 0:
overlap_indices.append([i, i + 1])
return overlap_indices
def exclude_artifact_words(self): '''Set words which overlap with artifact timeframe to usable == False, recounts nwords. sets the artifact id to the word object WIP: maybe seperate field for artifact usability on the word object ''' self.nallwords = self.nwords if self.artifacts == 'NA' or type(self.st_sample) != int: return 0 for w in self.words: for a in self.artifacts: o = utils.compute_overlap(w.st_sample - self.st_sample,w.et_sample - self.st_sample,a.st_sample,a.et_sample) if o > 0: w.set_artifact(a) break self.nwords = len([w for w in self.words if w.usable]) self.ncwords = len([w for w in self.words if w.usable and hasattr(w,'pos') and w.pos.content_word]) self.nexcluded = self.nallwords - self.nwords
def _generate_labels(self, visual_data: List[torch.Tensor], captions: List[Caption]):
"""
:param visual_data: a list of extracted features from videos
:param textual_data: list of annotations
:returns labels for the samples
"""
labels = []
for v, s in zip(visual_data, captions):
T = v.shape[0]
label = torch.zeros([T, self.K], dtype=torch.int32)
start_time, end_time = s.start_time, s.end_time
for t in range(T):
for k in range(self.K):
if (compute_overlap((t - (k+1) * self.delta) * self.sample_rate / self.fps,
t * self.sample_rate / self.fps,
start_time, end_time)) > self.threshold:
label[t, k] = 1
labels.append(label)
return pad_labels(labels) # torch.Tensor with shape (num_labels, T, K)
def evaluate(self,
generator,
iou_threshold=0.3,
score_threshold=0.3,
max_detections=100,
save_path=None):
""" Evaluate a given dataset using a given model.
code originally from https://github.com/fizyr/keras-retinanet
# Arguments
generator : The generator that represents the dataset to evaluate.
model : The model to evaluate.
iou_threshold : The threshold used to consider when a detection is positive or negative.
score_threshold : The score confidence threshold to use for detections.
max_detections : The maximum number of detections to use per image.
save_path : The path to save images with visualized detections to.
# Returns
A dict mapping class names to mAP scores.
"""
# gather all detections and annotations
all_detections = [[None for i in range(generator.num_classes())]
for j in range(generator.size())]
all_annotations = [[None for i in range(generator.num_classes())]
for j in range(generator.size())]
for i in range(generator.size()):
raw_image = generator.load_image(i)
raw_height, raw_width, raw_channels = raw_image.shape
# make the boxes and the labels
pred_boxes = self.predict(raw_image)
score = np.array([box.score for box in pred_boxes])
pred_labels = np.array([box.label for box in pred_boxes])
if len(pred_boxes) > 0:
pred_boxes = np.array([[
box.xmin * raw_width, box.ymin * raw_height,
box.xmax * raw_width, box.ymax * raw_height, box.score
] for box in pred_boxes])
else:
pred_boxes = np.array([[]])
# sort the boxes and the labels according to scores
score_sort = np.argsort(-score)
pred_labels = pred_labels[score_sort]
pred_boxes = pred_boxes[score_sort]
# copy detections to all_detections
for label in range(generator.num_classes()):
all_detections[i][label] = pred_boxes[pred_labels == label, :]
annotations = generator.load_annotation(i)
# copy detections to all_annotations
for label in range(generator.num_classes()):
all_annotations[i][label] = annotations[annotations[:, 4] ==
label, :4].copy()
# compute mAP by comparing all detections and all annotations
average_precisions = {}
for label in range(generator.num_classes()):
false_positives = np.zeros((0, ))
true_positives = np.zeros((0, ))
scores = np.zeros((0, ))
num_annotations = 0.0
for i in range(generator.size()):
detections = all_detections[i][label]
annotations = all_annotations[i][label]
num_annotations += annotations.shape[0]
detected_annotations = []
for d in detections:
scores = np.append(scores, d[4])
if annotations.shape[0] == 0:
false_positives = np.append(false_positives, 1)
true_positives = np.append(true_positives, 0)
continue
overlaps = compute_overlap(np.expand_dims(d, axis=0),
annotations)
assigned_annotation = np.argmax(overlaps, axis=1)
max_overlap = overlaps[0, assigned_annotation]
if max_overlap >= iou_threshold and assigned_annotation not in detected_annotations:
false_positives = np.append(false_positives, 0)
true_positives = np.append(true_positives, 1)
detected_annotations.append(assigned_annotation)
else:
false_positives = np.append(false_positives, 1)
true_positives = np.append(true_positives, 0)
# no annotations -> AP for this class is 0 (is this correct?)
if num_annotations == 0:
average_precisions[label] = 0
continue
# sort by score
indices = np.argsort(-scores)
false_positives = false_positives[indices]
true_positives = true_positives[indices]
# compute false positives and true positives
false_positives = np.cumsum(false_positives)
true_positives = np.cumsum(true_positives)
# compute recall and precision
recall = true_positives / num_annotations
precision = true_positives / np.maximum(
true_positives + false_positives,
np.finfo(np.float64).eps)
# compute average precision
average_precision = compute_ap(recall, precision)
average_precisions[label] = average_precision
return average_precisions
def evaluate(self):
self.yolo.model = self.model
# gather all detections and annotations
all_detections = [[
None for i in range(self.generator.num_classes())
] for j in range(self.generator.size())]
all_annotations = [[
None for i in range(self.generator.num_classes())
] for j in range(self.generator.size())]
for i in range(self.generator.size()):
raw_image = self.generator.load_image(i)
raw_height, raw_width, raw_channels = raw_image.shape
# make the boxes and the labels
pred_boxes = self.yolo.predict(raw_image)
score = np.array([box.score for box in pred_boxes])
pred_labels = np.array([box.label for box in pred_boxes])
if len(pred_boxes) > 0:
pred_boxes = np.array([[
box.xmin * raw_width, box.ymin * raw_height,
box.xmax * raw_width, box.ymax * raw_height, box.score
] for box in pred_boxes])
else:
pred_boxes = np.array([[]])
# sort the boxes and the labels according to scores
score_sort = np.argsort(-score)
pred_labels = pred_labels[score_sort]
pred_boxes = pred_boxes[score_sort]
# copy detections to all_detections
for label in range(self.generator.num_classes()):
all_detections[i][label] = pred_boxes[pred_labels ==
label, :]
annotations = self.generator.load_annotation(i)
# copy detections to all_annotations
for label in range(self.generator.num_classes()):
all_annotations[i][label] = annotations[
annotations[:, 4] == label, :4].copy()
# compute mAP by comparing all detections and all annotations
average_precisions = {}
for label in range(self.generator.num_classes()):
false_positives = np.zeros((0, ))
true_positives = np.zeros((0, ))
scores = np.zeros((0, ))
num_annotations = 0.0
for i in range(self.generator.size()):
detections = all_detections[i][label]
annotations = all_annotations[i][label]
num_annotations += annotations.shape[0]
detected_annotations = []
for d in detections:
scores = np.append(scores, d[4])
if annotations.shape[0] == 0:
false_positives = np.append(false_positives, 1)
true_positives = np.append(true_positives, 0)
continue
overlaps = compute_overlap(np.expand_dims(d, axis=0),
annotations)
assigned_annotation = np.argmax(overlaps, axis=1)
max_overlap = overlaps[0, assigned_annotation]
if max_overlap >= self.iou_threshold and assigned_annotation not in detected_annotations:
false_positives = np.append(false_positives, 0)
true_positives = np.append(true_positives, 1)
detected_annotations.append(assigned_annotation)
else:
false_positives = np.append(false_positives, 1)
true_positives = np.append(true_positives, 0)
# no annotations -> AP for this class is 0 (is this correct?)
if num_annotations == 0:
average_precisions[label] = 0
continue
# sort by score
indices = np.argsort(-scores)
false_positives = false_positives[indices]
true_positives = true_positives[indices]
# compute false positives and true positives
false_positives = np.cumsum(false_positives)
true_positives = np.cumsum(true_positives)
# compute recall and precision
recall = true_positives / num_annotations
precision = true_positives / np.maximum(
true_positives + false_positives,
np.finfo(np.float64).eps)
# compute average precision
average_precision = compute_ap(recall, precision)
average_precisions[label] = average_precision
return average_precisions
def evaluate(self,
model,
imgs,
obj_threshold=0.3,
nms_threshold=0.3,
iou_threshold=0.5):
"""
# Arguments
model : The model to evaluate.
imgs : list of parsed test_img dictionaries.
obj_threshold : The score confidence threshold to use for detections.
nms_threshold : The threshold used to consider when a detection is positive or negative.
# Returns
A dict mapping class names to mAP scores.
"""
# gather all detections and annotations
test_size = len(imgs)
all_detections = [[None for i in range(self.n_class)]
for j in range(test_size)]
all_annotations = [[None for i in range(self.n_class)]
for j in range(test_size)]
ious = []
for i in range(test_size):
image_name = imgs[i]['filename']
if '.jpg' not in image_name and '.png' not in image_name:
image_name += '.jpg'
raw_image = cv2.imread(image_name)
raw_height, raw_width, raw_channels = raw_image.shape
# make the boxes and the labels
pred_boxes = self.predict(model, raw_image, obj_threshold,
nms_threshold)
score = np.array([box.score for box in pred_boxes])
pred_labels = np.array([box.label for box in pred_boxes])
if len(pred_boxes) > 0:
pred_boxes = np.array([[
box.xmin * raw_width, box.ymin * raw_height,
box.xmax * raw_width, box.ymax * raw_height, box.score
] for box in pred_boxes])
else:
pred_boxes = np.array([[]])
# sort the boxes and the labels according to scores
score_sort = np.argsort(-score)
pred_labels = pred_labels[score_sort]
pred_boxes = pred_boxes[score_sort]
# copy detections to all_detections
for label in range(self.n_class):
all_detections[i][label] = pred_boxes[pred_labels == label, :]
annotations = load_annotation(imgs, i, self.labels)
# copy detections to all_annotations
for label in range(self.n_class):
all_annotations[i][label] = annotations[annotations[:, 4] ==
label, :4].copy()
# compute mAP by comparing all detections and all annotations
average_precisions = {}
for label in range(self.n_class):
false_positives = np.zeros((0, ))
true_positives = np.zeros((0, ))
scores = np.zeros((0, ))
num_annotations = 0.0
for i in range(test_size):
detections = all_detections[i][label]
annotations = all_annotations[i][label]
num_annotations += annotations.shape[0]
detected_annotations = []
for d in detections:
scores = np.append(scores, d[4])
if annotations.shape[0] == 0:
false_positives = np.append(false_positives, 1)
true_positives = np.append(true_positives, 0)
continue
overlaps = compute_overlap(np.expand_dims(d, axis=0),
annotations)
assigned_annotation = np.argmax(overlaps, axis=1)
max_overlap = overlaps[0, assigned_annotation]
ious.append(max_overlap)
if max_overlap >= iou_threshold and assigned_annotation not in detected_annotations:
false_positives = np.append(false_positives, 0)
true_positives = np.append(true_positives, 1)
detected_annotations.append(assigned_annotation)
else:
false_positives = np.append(false_positives, 1)
true_positives = np.append(true_positives, 0)
# no annotations -> AP for this class is 0 (is this correct?)
if num_annotations == 0:
average_precisions[label] = 0
continue
# sort by score
indices = np.argsort(-scores)
false_positives = false_positives[indices]
true_positives = true_positives[indices]
# compute false positives and true positives
false_positives = np.cumsum(false_positives)
true_positives = np.cumsum(true_positives)
# compute recall and precision
recall = true_positives / num_annotations
precision = true_positives / np.maximum(
true_positives + false_positives,
np.finfo(np.float64).eps)
# compute average precision
average_precision = compute_ap(recall, precision)
average_precisions[label] = average_precision
map_dict = {}
# print evaluation
for label, average_precision in average_precisions.items():
map_dict[self.labels[label]] = average_precision
print(self.labels[label], '{:.4f}'.format(average_precision))
print('mAP: {:.4f}'.format(
sum(average_precisions.values()) / len(average_precisions)))
print('average IOU: {:.4f}'.format(np.mean(ious)))
average_map = sum(
average_precisions.values()) / len(average_precisions)
return [average_map, map_dict, np.mean(ious)]
def evaluate(self,
generator,
iou_threshold=0.3,
score_threshold=0.3,
max_detections=100,
save_path=None):
""" Evaluate a given dataset using a given model.
code originally from https://github.com/fizyr/keras-retinanet
# Arguments
generator : The generator that represents the dataset to evaluate.
model : The model to evaluate.
iou_threshold : The threshold used to consider when a detection is positive or negative.
score_threshold : The score confidence threshold to use for detections.
max_detections : The maximum number of detections to use per image.
save_path : The path to save images with visualized detections to.
# Returns
A dict mapping class names to mAP scores.
"""
# gather all detections and annotations
all_detections = [[None for i in range(generator.num_classes())] for j in range(generator.size())]
all_annotations = [[None for i in range(generator.num_classes())] for j in range(generator.size())]
for i in range(generator.size()):
raw_image = generator.load_image(i)
raw_height, raw_width, raw_channels = raw_image.shape
# make the boxes and the labels
pred_boxes = self.predict(raw_image)
score = np.array([box.score for box in pred_boxes])
pred_labels = np.array([box.label for box in pred_boxes])
if len(pred_boxes) > 0:
pred_boxes = np.array([[box.xmin*raw_width, box.ymin*raw_height, box.xmax*raw_width, box.ymax*raw_height, box.score] for box in pred_boxes])
else:
pred_boxes = np.array([[]])
# sort the boxes and the labels according to scores
score_sort = np.argsort(-score)
pred_labels = pred_labels[score_sort]
pred_boxes = pred_boxes[score_sort]
# copy detections to all_detections
for label in range(generator.num_classes()):
all_detections[i][label] = pred_boxes[pred_labels == label, :]
annotations = generator.load_annotation(i)
# copy detections to all_annotations
for label in range(generator.num_classes()):
all_annotations[i][label] = annotations[annotations[:, 4] == label, :4].copy()
# compute mAP by comparing all detections and all annotations
average_precisions = {}
for label in range(generator.num_classes()):
false_positives = np.zeros((0,))
true_positives = np.zeros((0,))
scores = np.zeros((0,))
num_annotations = 0.0
for i in range(generator.size()):
detections = all_detections[i][label]
annotations = all_annotations[i][label]
num_annotations += annotations.shape[0]
detected_annotations = []
for d in detections:
scores = np.append(scores, d[4])
if annotations.shape[0] == 0:
false_positives = np.append(false_positives, 1)
true_positives = np.append(true_positives, 0)
continue
overlaps = compute_overlap(np.expand_dims(d, axis=0), annotations)
assigned_annotation = np.argmax(overlaps, axis=1)
max_overlap = overlaps[0, assigned_annotation]
if max_overlap >= iou_threshold and assigned_annotation not in detected_annotations:
false_positives = np.append(false_positives, 0)
true_positives = np.append(true_positives, 1)
detected_annotations.append(assigned_annotation)
else:
false_positives = np.append(false_positives, 1)
true_positives = np.append(true_positives, 0)
# no annotations -> AP for this class is 0 (is this correct?)
if num_annotations == 0:
average_precisions[label] = 0
continue
# sort by score
indices = np.argsort(-scores)
false_positives = false_positives[indices]
true_positives = true_positives[indices]
# compute false positives and true positives
false_positives = np.cumsum(false_positives)
true_positives = np.cumsum(true_positives)
# compute recall and precision
recall = true_positives / num_annotations
precision = true_positives / np.maximum(true_positives + false_positives, np.finfo(np.float64).eps)
# compute average precision
average_precision = compute_ap(recall, precision)
average_precisions[label] = average_precision
return average_precisions