label_size = cu_pixel
middle_size = cu_pixel * scale
input_cache = np.zeros([cache_size, cu_size * scale, cu_size * scale])
label_cache = np.zeros([cache_size, cu_size, cu_size])
cache_cnt = 0
####### Configuration ##########
fid = 0
h5er = h5Handler(h5_name)
x_up = math.floor(width / cu_size - 1)
y_up = math.floor(height / cu_size - 1)
####### Configuration ##########
print("Input size: ", input_size)
if not random_flag:
for i in range(f_cnt):
Y = read_frame(dec_name, i, height, width)
YY = read_frame(gt_name, i, height, width)
cv2.imwrite('dec.png', Y)
cv2.imwrite('gt.png', YY)
for lx in range(0, width, cu_size):
for ly in range(0, height, cu_size):
rx = lx + cu_size * scale
ry = ly + cu_size * scale
if rx >= width or ry >= height:
continue
# import IPython
# IPython.embed()
input_cache[cache_cnt, :, :] = Y[ly:ly + cu_size * scale,
lx:lx + cu_size * scale]
if mask_mean:
[f_id, y, x, mode] = line.split()
y = int(y)
x = int(x)
f_id = int(f_id)
mode = int(mode)
# --------------for debug------------------
#if f_id > 5:
# break
# --------------for debug------------------
if y == 0 and x == 0:
pc = 0
dc = 0
ac = 0
gt_img = read_frame(gt_path, f_id, height, width)
dec_img = read_frame(dec_path, f_id, height, width)
print(f_id)
if x == 0 or y == 0:
continue
# print([x, y])
input[:, :2048, :, :] = dec_img[x - block_size:x,
y - block_size:y + block_size].reshape(
[1, 2048, 1, 1]) / 255.0
input[:, 2048:, :, :] = dec_img[x:x + block_size,
y - block_size:y].reshape(
[1, 1024, 1, 1]) / 255.0
label[...] = gt_img[x:x + block_size, y:y + block_size].reshape(
[1, 1024, 1, 1]) / 255.0
# test_img[:32,:64] = input[:,:2048,:,:].reshape([32,64])
# test_img[32:,:32] = input[:,2048:,:,:].reshape([32,32])