def perturb_frame(frame,params):
dims = frame.image.get_hw()
rand_affine = RandomPerturber.generate_random_affine(dims/2,dims,params)
perturbed_frame = Frame(frame.image_path)
perturbed_frame.image = rand_affine.apply_to_image(frame.image,dims)
for i,obj in enumerate(frame.objects):
# filter out completely out of bound objects
perturbed_obj_box = rand_affine.apply_to_box(obj.box)
perturbed_polygons = rand_affine.apply_to_polygons(obj.polygons)
if Box.intersection(perturbed_obj_box,perturbed_frame.image.get_bounding_box()) is not None:
obj_copy = copy.deepcopy(obj)
obj_copy.box = perturbed_obj_box
obj_copy.polygons = perturbed_polygons
perturbed_frame.objects.append(obj_copy)
return perturbed_frame
def apply_affine_to_frame(frame, affine, output_size):
perturbed_frame = Frame(frame.image_path)
perturbed_frame.image = affine.apply_to_image(frame.image, output_size)
for i, obj in enumerate(frame.objects):
# filter out completely out of bound objects
perturbed_obj_box = affine.apply_to_box(obj.box)
perturbed_polygons = affine.apply_to_polygons(obj.polygons)
if Box.intersection(
perturbed_obj_box,
perturbed_frame.image.get_bounding_box()) is not None:
obj_copy = copy.deepcopy(obj)
obj_copy.box = perturbed_obj_box
obj_copy.polygons = perturbed_polygons
perturbed_frame.objects.append(obj_copy)
return perturbed_frame
def find_negative_crop(self, frame, objects):
# pick a random crop, check that it does not overlap with an existing target
# TODO, this is inefficient, fix this algorithm later
frame_size = frame.image.get_wh()
max_attempts = 10
for i in range(0, max_attempts):
randcx = random.randrange(self.crop_size[0] / 2,
frame_size[0] - self.crop_size[0] / 2)
randcy = random.randrange(self.crop_size[1] / 2,
frame_size[1] - self.crop_size[1] / 2)
new_box = Box(randcx - self.crop_size[0] / 2,
randcy - self.crop_size[1] / 2,
randcx + self.crop_size[0] / 2,
randcy + self.crop_size[1] / 2)
box_found = all(
Box.intersection(x.box, new_box) is None for x in objects)
if box_found:
return new_box
return None
def load_train(self):
frame1, frame2, neg_box, pos_box, anchor_box = None, None, None, None, None
# TODO, this should probably break if never find anything for a while
while neg_box is None:
indices = random.sample(self.frame_ids, 2)
frame1, frame2 = [self.source[x] for x in indices]
frame1_objs = filter(lambda x: x.obj_type in self.obj_types,
frame1.get_objects())
frame2_objs = filter(lambda x: x.obj_type in self.obj_types,
frame2.get_objects())
# get random pos boxes
pos_box = random.choice(frame1_objs).box
anchor_box = random.choice(frame2_objs).box
# find random neg crop
neg_box = self.find_negative_crop(frame1, frame1_objs)
perturbed_pos_box = RandomPerturber.perturb_crop_box(
pos_box, self.perturbations)
affine_crop0 = crop_image_resize(frame1.image, perturbed_pos_box,
self.crop_size)
pos_crop = affine_crop0.apply_to_image(frame1.image, self.crop_size)
affine_crop1 = crop_image_resize(frame2.image, anchor_box,
self.anchor_size)
anchor_crop = affine_crop1.apply_to_image(frame2.image,
self.anchor_size)
affine_crop2 = crop_image_resize(frame1.image, neg_box, self.crop_size)
neg_crop = affine_crop2.apply_to_image(frame1.image, self.crop_size)
# neg_crop.visualize(display=True,title='neg')
# now find all the boxes that intersect with the perturbed_pos_box
intersected_boxes = []
for obj in filter(lambda x: x.obj_type in self.obj_types,
frame1.get_objects()):
if Box.intersection(obj.box, perturbed_pos_box) is not None:
intersected_boxes.append(obj.box)
intersected_boxes = list(
map(lambda x: affine_crop0.apply_to_box(x), intersected_boxes))
# test display
# disp_frame = Frame.from_image_and_objects(pos_crop,[Object(box_crop)])
# disp_frame.visualize(display=True,title='pos frame')
# pos_crop.visualize(display=True,title='pos crop')
pos = torch.Tensor(
pos_crop.to_order_and_class(
Ordering.CHW, ValueClass.FLOAT01).get_data().astype(float))
neg = torch.Tensor(
neg_crop.to_order_and_class(
Ordering.CHW, ValueClass.FLOAT01).get_data().astype(float))
anchor = torch.Tensor(
anchor_crop.to_order_and_class(
Ordering.CHW, ValueClass.FLOAT01).get_data().astype(float))
# pos_map = generate_response_map_from_boxes(pos_crop.get_hw(),intersected_boxes)
# PTImage.from_2d_numpy(pos_map).visualize(display=True,title='pos frame')
pos_map = torch.Tensor(
generate_response_map_from_boxes(pos_crop.get_hw(),
intersected_boxes))
neg_map = torch.Tensor(
generate_response_map_from_boxes(pos_crop.get_hw()))
data = [pos, neg, anchor]
target = [pos_map, neg_map, anchor]
return TripletDetectionSample(data, target)
