def prepare_tracking(srcRoot, gtfile, videoname, imageSz):
try:
gtboxes = np.loadtxt(gtfile, delimiter=',') # x y w h
except:
gtboxes = np.loadtxt(gtfile)
if videoname == 'Tiger1': # in OTB, for 'Tiger1', the starting frame is the 6th
gtboxes = gtboxes[5:, :]
firstframename = '0001.jpg'
if videoname == 'David':
firstframename = '0300.jpg'
if videoname == 'Tiger1': ##################################
firstframename = '0006.jpg'
img_path = srcRoot + videoname + '/' + 'img/' + firstframename
firstframe = image_io.load_image(img_path)
init_box = gtboxes[0, :].copy() # x y w h
nr_frames2track = gtboxes.shape[0] # including first frame
frameidx = range(2, nr_frames2track + 1)
if videoname == 'David':
frameidx = range(301, nr_frames2track + 300)
if videoname == 'Tiger1':
frameidx = range(7, nr_frames2track + 6)
input_roi = np.zeros((1, 5))
input_roi[0, 1] = init_box[0] * imageSz / firstframe.shape[1]
input_roi[0, 2] = init_box[1] * imageSz / firstframe.shape[0]
input_roi[0, 3] = (init_box[0] + init_box[2] -
1) * imageSz / firstframe.shape[1]
input_roi[0, 4] = (init_box[1] + init_box[3] -
1) * imageSz / firstframe.shape[0]
input_roi[0, 1:] -= 1 # starting from 0
input_roi = input_roi[np.newaxis, :]
return firstframe, input_roi, init_box, frameidx, nr_frames2track
def create_mask(image_path: hug.types.text,
algorithm: image_mask_algorithm_type,
morph: hug.types.number = 1):
"""Runs the masking algorithm and returns a serialized masked image."""
image_rgb = load_image(image_path)
mask = algorithm(image_rgb)
if morph > 0:
mask = apply_morphology(mask, iterations=morph)
image_rgb[mask == 0] = MASK_COLOR
serialized = serialize_image(image_rgb, OUTPUT_IMAGE_FORMAT)
return {
'type': 'image/{}'.format(OUTPUT_IMAGE_FORMAT),
'encoding': 'base64',
'content': serialized
}
def get_clip(clip, verbose=False):
'''
:param clip_name: str, the name of the clip (subfolder in c3d-pytorch/data)
:param verbose: if True, show the unrolled clip
:return:
'''
img_mean = np.asarray([128])
imageSz = (128, 171)
crop_size = 112
# clip = sorted(glob(os.path.join('./c3d-pytorch/data', clip_name, '*.jpg')))
clip = np.array([
preproc_img(image_io.load_image(frame),
mean=img_mean,
imageSz=imageSz,
crop_size=crop_size) for frame in clip
])
clip = clip.transpose(1, 0, 2, 3)
clip = clip[np.newaxis, :, :, :, :]
return torch.from_numpy(clip)
gtboxes = np.loadtxt(gtfile, delimiter=',') # x y w h
except:
gtboxes = np.loadtxt(gtfile)
if videoname == 'Tiger1': # in OTB, for 'Tiger1', the starting frame is the 6th
gtboxes = gtboxes[5:, :]
firstframename = '0001.jpg'
if videoname == 'David':
firstframename = '0300.jpg'
if videoname == 'Tiger1': ##################################
firstframename = '0006.jpg'
img_path = srcRoot + videoname + '/' + 'img/' + firstframename
firstframe = image_io.load_image(img_path)
init_box = gtboxes[0, :].copy() # x y w h
nr_frames2track = gtboxes.shape[0] # including first frame
input_roi = np.zeros((1, 5))
input_roi[0, 1] = init_box[0] * imageSz / firstframe.shape[1]
input_roi[0, 2] = init_box[1] * imageSz / firstframe.shape[0]
input_roi[0, 3] = (init_box[0] + init_box[2] -
1) * imageSz / firstframe.shape[1]
input_roi[0, 4] = (init_box[1] + init_box[3] -
1) * imageSz / firstframe.shape[0]
input_roi[0, 1:] -= 1 # starting from 0
input_roi = input_roi[np.newaxis, :]
model.bind(for_training=False,
data_shapes=[('data', (1, 3, imageSz, imageSz)),