def time(in_path, out_path):
def pairs(seq_i):
return [
np.concatenate([seq_i[j], seq_i[j + 1]])
for j in range(len(seq_i) - 1)
]
imgs.transform(in_path, out_path, pairs, single_frame=False)
def sigma_filtr(in_path, out_path):
def helper(img_i):
std_z = np.std(img_i)
value_i = np.mean(img_i) + 0.5 * std_z
print(value_i)
img_i[img_i < value_i] = 0
return img_i
imgs.transform(in_path, out_path, helper, True)
def time(in_path, out_path):
def helper(frames):
size = len(frames)
return [
np.concatenate([frames[i - 1], frames[i]], axis=0)
for i in range(1, size)
]
transform = [proj.scale, helper]
imgs.transform(in_path, out_path, transform, single_frame=False)
def projection(in_path, out_path):
proj_x = get_projection(dim=0)
proj_y = get_projection(dim=1)
def full_projection(frames):
x_frames = proj_x(frames)
y_frames = proj_y(frames)
return [
np.concatenate([proj_x, proj_y], axis=0)
for proj_x, proj_y in zip(x_frames, y_frames)
]
imgs.transform(in_path, out_path, full_projection, False)
def transform(in_path, out_path, type="diff"):
single = True
if (type == "diff"):
fun = diff_helper
single = False
if (type == "motion"):
fun = motion_helper
single = False
if (type == "canny"):
fun = lambda img_i: cv2.Canny(img_i, 100, 200)
if (type == "smooth"):
fun = smooth
if (type == "noise"):
fun = lambda img_i: np.abs(img_i - smooth(img_i))
imgs.transform(in_path, out_path, fun, single)
def reconstruct(in_path, model_path, out_path=None, diff=False):
model = load_model(model_path)
if (not out_path):
out_path = os.path.split(in_path)[0] + '/rec'
def rec_helper(X):
X = np.array(X)
X = data.format_frames(X)
pred = model.predict(X)
if (diff):
pred = np.abs(pred - X)
pred = [np.vstack(frame_i.T) for frame_i in pred]
return pred
imgs.transform(in_path, out_path, rec_helper, False)
def sample_imgs(in_path,out_path):
imgs.transform(in_path,out_path,sample_seq)
def raw_projection(in_path, out_path, dim=0):
helper = get_helper(dim)
imgs.transform(in_path, out_path, helper, False)
def smooth_frames(in_path, out_path):
fun = [gauss_helper, scale]
imgs.transform(in_path, out_path, gauss_helper, single_frame=True)
def scaled_frames(in_path, out_path):
imgs.transform(in_path, out_path, scale, single_frame=True)
def box_frame(in_path, out_path):
fun = [balanced_frames]
imgs.transform(in_path, out_path, fun, single_frame=False)
def rescale_imgs(in_path, out_path):
imgs.transform(in_path, out_path, norm_z, single_frame=False)
def box_frame(in_path, out_path):
fun = [equal_box] #[extract_box]
imgs.transform(in_path, out_path, fun, single_frame=False)
def box_frame(in_path, out_path):
imgs.transform(in_path, out_path, extract_box, single_frame=False)
def rescale_imgs(in_path, out_path, dim_x=64, dim_y=128):
rescale = proj.Scale(dim_x, dim_y)
if (files.dict_of_dicts(in_path)):
imgs.transform(in_path, out_path, rescale, True)
else:
imgs.transform_action_img(in_path, out_path, rescale)
def rescale(in_path, out_path, dim_x=64, dim_y=64):
imgs.transform(in_path, out_path, scale, single_frame=True)
for point_i in points.T:
if(point_i[0]>1.5*self.center[0]):
dist=np.abs(point_i[0]-self.center[0])
dist/=self.center[0]
delta=dist*self.scale
if(point_i[1]<self.center[1]):
point_i[1]-=delta
else:
point_i[1]+=delta
return points
def gap_agum(frames):
frames=[ proj.nonzero_points(frame_i)
for frame_i in frames]
center=pclouds.center_of_mass(frames)
# frames=[filtr_points(frame_i,1.5*center[0]) for frame_i in frames]
helper=GapTransform(center,16.0)
frames=[helper(frame_i) for frame_i in frames]
frames=[ pclouds.to_img(frame_i) for frame_i in frames]
return frames
def filtr_points(points,threshold):
new_points=[]
for point_i in points.T:
if(point_i[0]<threshold):
new_points.append(point_i)
return np.array(new_points).T
if __name__ == "__main__":
imgs.transform("../agum/box","test",gap_agum)
def outliner_img(in_path,out_path):
fun=[tools.median_smooth,outliner,
proj_center.center_norm,
pclouds.to_img,
proj.scale]
imgs.transform(in_path,out_path,fun)
def imgs_only(in_path, out_path):
fun = imgs.Pipeline([center_pcloud, pclouds.to_img])
imgs.transform(in_path, out_path, fun)
