def plot_moving_inputs(inputs, labels, opts):
rc = (2, 3)
fig, ax = plt.subplots(rc[0], rc[1])
state = [x[:, :opts.state_size] for x in inputs[:rc[0]]]
extra = [x[:, opts.state_size:opts.state_size + 2] for x in inputs[:rc[0]]]
labels = labels[:rc[0]]
i = 0
for batch in zip(state, extra, labels):
for d in batch:
plot_ix = np.unravel_index(i, rc)
cur_ax = ax[plot_ix]
adjust(cur_ax)
plt.sca(cur_ax)
plt.imshow(d, cmap='RdBu_r', vmin=-.3, vmax=.3)
if i % 3 != 1:
plt.xticks([0, 19])
else:
plt.xticks([])
plt.yticks([0, 49])
cb = plt.colorbar()
cb.set_ticks([0, .3])
i += 1
path = os.path.join('./lab_meeting/images', 'input_non_stationary')
plt.savefig(path + '.png', dpi=300)
def test_booleans(test_case):
print("Running", "test_booleans")
test_case.runAsJava(adjust("""
for i in range(100000):
x = True
y = False
x = (not x and y)
y = x != y
x = True and True and True
y = x or False
x = not True or (not x) or True
y = y == y
x = y != (not x)
y = False or True or (not y)
x = (x != x) or (not False)
y = y and y and (not x)
x = False or False or False or False
y = True and True and True and True
x = 1 != 0
y = (not (not (not (not (not (not (not x)))))))
x = ((not y) and y) == ((not (not x)) or y or (not y))
y = False or False or (not True) or (not True) or (not True)
x = True and (not False) and True and (not False) and False
y = (False != False) != False
"""),
timed=True)
def test_cmp(test_case):
print("Running", "test_cmp")
test_case.runAsJava(adjust("""
x = None
s = "mary had a little lamb"
t = "humpty dumpty sat on a wall"
for i in range(1000):
for j in range(1000):
x = s < t
x = s <= t
x = s == t
x = s != t
x = s > t
x = s >= t
x = 3 < 5
x = 3 <= 5
x = 3 == 5
x = 3 != 5
x = 3 > 5
x = 3 >= 5
x = 3 < True
x = 3.0 <= 5
x = None == 5
x = [3] != 5.0
x = [3] > [5]
x = [3.0] > [5.0]
"""),
timed=True)
def test_code(test_case):
print("Running", "test_code")
test_case.runAsJava(adjust("""
def main(n):
def foo(n):
return n + 1
def bar(n):
return n + 1
def baz(n):
return n + 1
def buzz(n):
return n + 1
def bizz(n):
return n + 1
def bozz(n):
return n + 1
def yam(n):
return n + 1
def yarn(n):
return n + 1
return foo(bar(baz(buzz(bizz(bozz(yam(yarn(n))))))))
for i in range(100000):
main(i)
"""),
timed=True)
def test_loops(test_case):
print("Running", "test_loops")
test_case.runAsJava(adjust("""
for x in range(100):
for y in range(100):
for z in range(100):
for a in range(100):
pass
"""),
timed=True)
def test_small_integers(test_case):
print("Running", "test_small_integers")
test_case.runAsJava(adjust("""
for i in range(100):
for j in range(100):
x = -5
for k in range(-5, 257):
x = x + 1
"""),
timed=True)
def plot_stationary_inputs(inputs, labels, opts):
rc = (2, 2)
fig, ax = plt.subplots(rc[0], rc[1])
state = inputs[:rc[0]]
labels = labels[:rc[0]]
i = 0
for batch in zip(state, labels):
for d in batch:
plot_ix = np.unravel_index(i, rc)
cur_ax = ax[plot_ix]
adjust(cur_ax)
plt.sca(cur_ax)
# cbarBoo = True if i %2==1 else True
plt.imshow(d, cmap='RdBu_r', vmin=-.3, vmax=.3)
plt.xticks([0, 19])
plt.yticks([0, 49])
cb = plt.colorbar()
cb.set_ticks([0, .3])
i += 1
path = os.path.join('./lab_meeting/images', 'input_stationary')
plt.savefig(path + '.png', dpi=300)
def test_dict_get(test_case):
print("Running", "test_dictionary_get")
test_case.runAsJava(adjust("""
dict = {1 : 2, "a" : "b"}
for i in range(1000000):
dict.get(i)
dict.get(1)
dict.get("a")
dict.get(i)
dict.get(1)
dict.get("a")
"""),
timed=True)
def test_dict_set(test_case):
print("Running", "test_dictionary_set")
test_case.runAsJava(adjust("""
dict = {}
for i in range(1000000):
dict["moo"] = 1
dict["quack"] = 2
dict["woof"] = 3
dict["meow"] = 4
dict["cockadoodledoo"] = 5
dict["hiss"] = 6
"""),
timed=True)
def adjust(self, method, frames=None, inplace=True, **kwargs):
if frames is None and len(self.frames) == 0:
raise ValueError("You did not specify any frames")
if frames is not None:
frames = frames
else:
frames = self.frames
adjusted = []
for frame in frames:
adjusted.append(utils.adjust(frame, method, **kwargs))
if inplace:
self.frames = adjusted
else:
return adjusted
def create_normals(normal_map):
h, w, _ = normal_map.shape
iterations = w * h
step_size = np.ceil(iterations / 100).astype(int)
normals = np.zeros((h, w, NORMAL_DIMENSIONS))
counter = 0
bar = Bar("Processing Normals...", max=100, suffix='%(percent)d%%')
bar.check_tty = False
for i in range(w):
for j in range(h):
normals[j][i] = utils.adjust(normal_map[j][i][:3])
counter += 1
if counter % step_size == 0:
bar.next()
bar.finish()
np.save(NORMAL_VECTORS_FILENAME, normals)
print(f"Normal vectors stored in {NORMAL_VECTORS_FILENAME}")
return normals
def test_global_var_load(test_case):
print("Running", "test_global_var_load")
test_case.runAsJava(adjust("""
x = 1
y = 2
for i in range(100000):
print(x)
print(y)
print(x+y)
print(x-y)
print(x)
print(y)
print(x*y)
print(x/y)
print(x)
print(y)
"""),
timed=True)
def test_function_var_load(test_case):
print("Running", "test_function_var_load")
test_case.runAsJava(adjust("""
def foo():
x = 1
y = 2
print(x)
print(y)
print(x+y)
print(x-y)
print(x)
print(y)
print(x*y)
print(x/y)
print(x)
print(y)
for i in range(100000):
foo()
"""),
timed=True)
def test_method(test_case):
print("Running", "test_method")
test_case.runAsJava(adjust("""
class MyClass:
def A(self):
print("A!")
def B(self):
print("B!")
def C(self):
print("C!")
obj = MyClass()
for i in range(1000000):
obj.A()
obj.B()
obj.C()
"""),
timed=True)
def test_class_init(test_case):
print("Running", "test_class_init")
test_case.runAsJava(adjust("""
class A: pass
class B: pass
class C: pass
class D: pass
class E: pass
class F: pass
class G: pass
class H: pass
class I: pass
class J: pass
class K: pass
class L: pass
class M: pass
class N: pass
class O: pass
class P: pass
class Q: pass
class R: pass
class S: pass
class T: pass
class U: pass
class V: pass
class W: pass
class X: pass
class Y: pass
class Z: pass
a = "a"
b = 1
c = [1]
d = {1 : 1}
e = list([1])
f = True
g = 3.0
"""),
timed=True)
def adjust_normal_map(rgb_normal_map):
"""
Return an adjusted normal map on which each element is a normalized vector
from a RGB normal map.
Args:
rgb_normal_map(numpy.array): The RGB normal map as a numpy array.
Returns:
numpy.array: Map of normalized vector normals.
"""
print("Creating normal vectors from RGB map...")
h, w, _ = rgb_normal_map.shape
iterations = w * h
step_size = np.ceil((iterations * PERCENTAGE_STEP) / 100).astype('int')
normals = np.zeros((h, w, NORMAL_DIMENSIONS))
counter = 0
for i in range(w):
for j in range(h):
if counter % step_size == 0:
percent_done = int((counter / float(iterations)) * 100)
print("{}% of normal vectors created".format(percent_done))
normals[j][i] = utils.adjust(rgb_normal_map[j][i][:3])
counter += 1
return normals
def test_class_var_load(test_case):
print("Running", "test_class_var_load")
test_case.runAsJava(adjust("""
class Animal:
def __init__(self, name, sound):
self.name = name
self.sound = sound
def speak(self):
print(self.name)
print(self.sound)
print(self.name)
print(self.sound)
print(self.name)
print(self.sound)
lapo = Animal("Lapo", "Bow Wow")
for i in range(100000):
lapo.speak()
lapo.speak()
lapo.speak()
"""),
timed=True)
if args.cuda:
net.load_state_dict(copyStateDict(torch.load(args.trained_model)))
else:
net.load_state_dict(copyStateDict(torch.load(args.trained_model, map_location='cpu')))
if args.cuda:
net = net.cuda()
net = torch.nn.DataParallel(net)
cudnn.benchmark = False
net.eval()
t = time.time()
# load data
for k, image_path in enumerate(image_list):
print("Test image {:d}/{:d}: {:s}".format(k+1, len(image_list), image_path), 'end=\r')
image = imgproc.loadImage(image_path)
bboxes, polys, score_text = test_net(net, image, args.text_threshold, args.link_threshold, args.low_text, args.cuda, args.poly)
# save score text
filename, file_ext = os.path.splitext(os.path.basename(image_path))
mask_file = result_folder + "/res_" + filename + '_mask.jpg'
cv2.imwrite(mask_file, score_text)
polys = adjust(polys)
file_utils.saveResult(image_path, image[:, :, ::-1], polys, dirname=result_folder)
print("elapsed time : {}s".format(time.time() - t))
def forward(self, x, std_val=0.01, constant_rep=True):
#create the latent vector
#-----------------------------------------------------------------------
x = self.encoder(x)
#-----------------------------------------------------------------------
if constant_rep:
N = self.num_points / self.nb_primitives
ones = Variable(torch.ones(x.size(0), self.nb_primitives)).cuda()
points_per_primitive = ones.contiguous() * N
points_per_primitive = points_per_primitive.int()
existance_prob = ones[1]
else:
#compute the means for the normal distributions
#-------------------------------------------------------------------
mean = self.existanceProbability(x.unsqueeze(2))
mean = mean.view(x.size(0), self.nb_primitives).contiguous()
#-------------------------------------------------------------------
#fixed standart deviation for every normal distriutions
#-------------------------------------------------------------------
ones = Variable(torch.ones(x.size(0), self.nb_primitives)).cuda()
stddev = (std_val * ones).contiguous()
#-------------------------------------------------------------------
#sample over the normal distriutions the existance probabilities and
#make sure they are in [0,1]
#-------------------------------------------------------------------
existance_prob = torch.normal(mean, stddev).contiguous()
existance_prob[existance_prob < 0] = 0
existance_prob[existance_prob > 1] = 1
# existance_prob_ = existance_prob
#-------------------------------------------------------------------
# normalize the probabilities for every batch
#-------------------------------------------------------------------
batch_sum_prob = torch.sum(existance_prob, 1).contiguous()
batch_sum_prob = batch_sum_prob.expand(self.nb_primitives,
x.size(0))
batch_sum_prob = batch_sum_prob.permute(1, 0)
batch_sum_prob = batch_sum_prob.contiguous()
existance_prob = torch.div(existance_prob, batch_sum_prob)
existance_prob = existance_prob.contiguous()
#-------------------------------------------------------------------
#computing the number of points per primitives wrt the batch
#-------------------------------------------------------------------
points_per_primitive = torch.round(existance_prob *
self.num_points)
points_per_primitive = points_per_primitive.int().contiguous()
points_per_primitive = adjust(points_per_primitive,
self.num_points)
#-------------------------------------------------------------------
#create on spatial transformation per primitive
#-----------------------------------------------------------------------
linear_list = []
for primitive in range(self.nb_primitives):
linear_list.append(self.rotBias[primitive](
x.unsqueeze(2).contiguous()))
linear_list = torch.cat(linear_list, 2)
#-----------------------------------------------------------------------
out_batch = []
for batch in range(x.size(0)):
out_primitive = []
for primitive in range(self.nb_primitives):
#find the number of points to generate wrt the batch and the
#primitive number
#---------------------------------------------------------------
N = points_per_primitive[batch, primitive].data[0]
#---------------------------------------------------------------
#ignore the primitive is there is no point to associated
#---------------------------------------------------------------
if (N == 0):
continue
#---------------------------------------------------------------
#generete the 2D-primitive wrt the number of points
#---------------------------------------------------------------
rand_grid = Variable(torch.cuda.FloatTensor(1, 2, N))
rand_grid.data.uniform_(0, 1)
#---------------------------------------------------------------
#allowing the network to modify the plan before the spatial
#transformation
#---------------------------------------------------------------
if self.D3_is_on:
#tranform the 2D primitive into a 3D surface
#-----------------------------------------------------------
y = self.d2Tod3[primitive](rand_grid)
#-----------------------------------------------------------
#generate the roation matrix & bias wrt the primitive
#-----------------------------------------------------------
linear = linear_list[batch, :, primitive]
q = linear[0:9]
q = q.view(3, 3)
q = q.contiguous()
t = linear[9:12]
t = t.view(1, 3).expand(y.size(2), -1, -1).permute(1, 2, 0)
t = t.contiguous()
#-----------------------------------------------------------
#---------------------------------------------------------------
#not allowing the network to modify the plan before the spatial
#transformation (ie using simple plan)
#---------------------------------------------------------------
else:
y = rand_grid
#generate the roation matrix & bias wrt the primitive
#-----------------------------------------------------------
linear = linear_list[batch, :, primitive]
q = linear[0:6]
q = q.view(2, 3)
q = q.contiguous()
t = linear[6:9]
t = t.view(1, 3)
t = t.expand(y.size(2), -1, -1).permute(1, 2, 0)
t = t.contiguous()
#-----------------------------------------------------------
#---------------------------------------------------------------
#apply the transformation matrix and the bias
#---------------------------------------------------------------
yq = torch.matmul(y.permute(0, 2, 1).contiguous(), q)
yq = yq.permute(0, 2, 1).contiguous()
yqt = torch.add(yq, t)
#---------------------------------------------------------------
#combine all the predictions of the primitives of one sample
#---------------------------------------------------------------
out_primitive.append(yqt)
out_batch.append(torch.cat(out_primitive, 2).contiguous())
#-------------------------------------------------------------------
#combine all the predictions of the sample
#-----------------------------------------------------------------------
out_batch = torch.cat(out_batch, 0).contiguous()
#-----------------------------------------------------------------------
#configuration and its probability
#-----------------------------------------------------------------------
existance_prob[existance_prob == 0] = 1
configuration = out_batch.transpose(2, 1).contiguous()
if constant_rep:
ones = Variable(torch.ones(x.size(0))).cuda().contiguous()
configuration_probability = ones
else:
configuration_probability = torch.prod(existance_prob, 1)
#-----------------------------------------------------------------------
return configuration, configuration_probability, points_per_primitive
