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, copy.deepcopy(frame.objects))
perturbed_frame.image = rand_affine.apply_to_image(frame.image, dims)
for i, obj in enumerate(frame.objects):
perturbed_frame.objects[i].box = rand_affine.apply_to_box(obj.box)
return perturbed_frame
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 __getitem__(self, index):
image, labels = self.dataset[index]
np_arr = np.asarray(image)
ptimage = PTImage.from_numpy_array(np_arr)
objects = []
for t in labels:
box = Box.from_xywh(t['bbox'])
obj_type = self.coco.loadCats([t['category_id']])[0]['name']
# convert segmentation to polygon using the pycocotools
# note the segmentation could in one of several formats, for example the custom coco RLE,
# to convert the RLE back to polygon is bit of a pain so I will just ignore those right now
# according the COCO site, most of the data is in polygon form (not sure why theres a discrepency?)
# and I'd rather not store 2D binary masks with every object.
polygon = t.get('segmentation')
# reshape to 2d poly, assume its convex hull?
polys = []
if polygon and isinstance(polygon, list):
for seg in polygon:
polys.append(
Polygon(
np.array(seg).reshape((int(old_div(len(seg),
2)), 2))))
objects.append(Object(box, obj_type=obj_type, polygons=polys))
frame = Frame.from_image_and_objects(ptimage, objects)
return frame
def __getitem__(self,index):
pil_img,label = self.dataset[index]
# assert 2D here
np_arr = np.asarray(pil_img)
np_arr = np.expand_dims(np_arr, axis=2)
# create the PTImage, and object that span the frame
# add extra channel dimension
ptimage = PTImage.from_numpy_array(np_arr)
obj = Object(Box(0,0,pil_img.size[0],pil_img.size[1]))
frame = Frame.from_image_and_objects(ptimage,[obj])
return frame
def __load_labelled_frames(self, frame_dir, labels_file):
files = [f for f in listdir(frame_dir) if isfile(join(frame_dir, f))]
labels = KITTILabel.labels_from_file(
labels_file) if labels_file is not None else []
frames = []
for f in files:
file_index = int(splitext(basename(f))[0])
objects = []
for l in labels:
if int(l.frame_idx) == file_index:
objects.append(l.to_object())
frames.append(Frame(join(frame_dir, f), objects))
return frames
def __load_labelled_frames(self,frame_dir,labels_file,calib_file=None):
files = [f for f in listdir(frame_dir) if isfile(join(frame_dir, f))]
labels = KITTILabel.labels_from_file(labels_file) if labels_file is not None else []
calibration_mat = None
if os.path.isfile(calib_file):
calibration_mat = self.__load_camera_matrix_from_calib(calib_file)
frames = []
sorted_files = sorted(files, key=lambda x: int(splitext(basename(x))[0]))
for f in sorted_files:
file_index = int(splitext(basename(f))[0])
objects = []
for l in labels:
if int(l.frame_idx) == file_index:
objects.append(l.to_object())
frames.append(Frame(join(frame_dir,f),objs=objects,calib_mat=calibration_mat))
return frames
def __load_frames(self, cars_dir, labels_mat):
print('Loading Stanford Cars Frames')
labels = scipy.io.loadmat(labels_mat)['annotations'][0]
# load frames with labels
for label in labels:
if len(label) == 5:
xmin, ymin, xmax, ymax, path = label
elif len(label) == 6:
xmin, ymin, xmax, ymax, _, path = label
else:
assert False, 'unable to parse label!'
box = Box(float(xmin[0][0]), float(ymin[0][0]), float(xmax[0][0]),
float(ymax[0][0]))
obj = Object(box, obj_type='car')
image_path = os.path.join(cars_dir, path[0])
self.frames.append(Frame(image_path, [obj]))