def use_model(self, x, t):
self.reset()
num_lm = x.shape[0]
n_step = self.n_step
s_list = xp.empty((n_step, num_lm, 1))
l_list = xp.empty((n_step, num_lm, 2))
x_list = xp.empty((n_step, num_lm, 3, self.gsize, self.gsize))
l, s, b1 = self.first_forward(x, num_lm)
for i in range(n_step):
if i + 1 == n_step:
xm, lm, sm = self.make_img(x, l, s, num_lm, random=0)
l1, s1, y, b = self.recurrent_forward(xm, lm, sm)
s_list[i] = sm.data
l_list[i] = lm.data
x_list[i] = xm.data
accuracy = y.data * t
s_list = xp.power(10, s_list - 1)
return xp.sum(accuracy, axis=1), l_list, s_list, x_list
else:
xm, lm, sm = self.make_img(x, l, s, num_lm, random=0)
l1, s1, y, b = self.recurrent_forward(xm, lm, sm)
l = l1
s = s1
s_list[i] = sm.data
l_list[i] = lm.data
x_list[i] = xm.data
return
def generate_xm_in_gpu(lm, sm, img, num_lm, g_size, img_size=112):
xm = xp.empty((num_lm, g_size * g_size)).astype(xp.float32)
img_buf = img.reshape((num_lm, img_size * img_size))
zm = xp.power(10, sm - 1)
for k in range(num_lm):
xr = xp.linspace((lm[k][0] - zm[k] / 2), (lm[k][0] + zm[k] / 2),
g_size)
xr *= img_size
xr = xp.clip(xr, 0, img_size - 1).astype(np.int32)
yr = xp.linspace((lm[k][1] - zm[k] / 2), (lm[k][1] + zm[k] / 2),
g_size)
yr *= img_size
yr = xp.clip(yr, 0, img_size - 1).astype(np.int32)
xr = img_size * np.repeat(xr, g_size) + xp.tile(yr, g_size)
xm[k] = img_buf[k][xr]
return xm.reshape(num_lm, 1, g_size, g_size).astype(xp.float32)
def use_model(self, x, t):
self.reset()
num_lm = x.shape[0]
n_step = self.n_step
s_list = xp.ones((n_step, num_lm, 1)) * (self.gsize / self.img_size)
l_list = xp.empty((n_step, num_lm, 2))
l, b1 = self.first_forward(x, num_lm)
for i in range(n_step):
if i + 1 == n_step:
xm, lm = self.make_img(x, l, num_lm, random=0)
l1, y, b = self.recurrent_forward(xm, lm)
l_list[i] = l1.data
accuracy = y.data * t
return xp.sum(accuracy, axis=1), l_list, s_list
else:
xm, lm = self.make_img(x, l, num_lm, random=0)
l1, y, b = self.recurrent_forward(xm, lm)
l = l1
l_list[i] = l.data
return