def discrim_l2(self, y, y_, step):
"""
facial semantic feature loss
evaluate loss use l1 at pixel space
:param y: input photo, numpy array [H, W, C]
:param y_: generated image, tensor [B, C, W, H]
:param step: train step
:return: l1 loss in pixel space
"""
img1 = parse_evaluate(y.astype(np.uint8),
cp=self.parsing,
cuda=self.cuda)
y_ = y_.cpu().detach().numpy()
y_ = np.squeeze(y_, axis=0)
y_ = np.swapaxes(y_, 0, 2) * 255
img2 = parse_evaluate(y_.astype(np.uint8),
cp=self.parsing,
cuda=self.cuda)
edge_img1 = utils.img_edge(img1).astype(np.float32)
edge_img2 = utils.img_edge(img2).astype(np.float32)
w_g = 1.0
w_r = 1.0
if step % self.args.eval_prev_freq == 0:
path = os.path.join(self.prev_path, "l2_{0}.jpg".format(step))
edge1_v3 = 255. - ops.fill_grey(edge_img1)
edge2_v3 = 255. - ops.fill_grey(edge_img2)
merge = ops.merge_4image(y,
y_,
edge1_v3,
edge2_v3,
transpose=False)
cv2.imwrite(path, merge)
return np.mean(np.abs(w_r * edge_img1 - w_g * edge_img2))
def capture(path, tensor1, tensor2, parse, cuda):
"""
imitator 快照
:param cuda: use gpu
:param path: save path
:param tensor1: input photo
:param tensor2: generated image
:param parse: parse checkpoint's path
"""
img1 = ops.tensor_2_image(tensor1)[0].swapaxes(0, 1).astype(np.uint8)
img2 = ops.tensor_2_image(tensor2)[0].swapaxes(0, 1).astype(np.uint8)
img1 = cv2.resize(img1, (512, 512), interpolation=cv2.INTER_LINEAR)
img3 = utils.faceparsing_ndarray(img1, parse, cuda)
img4 = utils.img_edge(img3)
img4 = 255 - ops.fill_gray(img4)
image = ops.merge_4image(img1, img2, img3, img4, transpose=False)
cv2.imwrite(path, image)
def output(self, x, refer, step):
"""
capture for result
:param x: generated image with grad, torch tensor [b,params]
:param refer: reference picture
:param step: train step
"""
self.write(x)
y_ = self.imitator(x)
y_ = y_.cpu().detach().numpy()
y_ = np.squeeze(y_, axis=0)
y_ = np.swapaxes(y_, 0, 2) * 255
y_ = y_.astype(np.uint8)
im1 = self.l2_c[0]
im2 = self.l2_c[1]
np_im1 = im1.cpu().detach().numpy()
np_im2 = im2.cpu().detach().numpy()
f_im1 = ops.fill_gray(np_im1)
f_im2 = ops.fill_gray(np_im2)
image_ = ops.merge_4image(refer, y_, f_im1, f_im2, transpose=False)
path = os.path.join(self.prev_path, "eval_{0}.jpg".format(step))
cv2.imwrite(path, image_)
def capture(self, tensor1, tensor2, name, step, cuda):
"""
extractor 快照
:param tensor1: input photo
:param tensor2: generated image
:param cuda: use gpu to speed up
:param step: train step
:param name: picture name
"""
path = "{1}/{2}_{0}.jpg".format(step, self.prev_path, name[3:-6])
orig_path = os.path.join(self.args.path_to_dataset + "2", name)
img3 = cv2.imread(orig_path)
img4 = utils.faceparsing_ndarray(img3, self.args.parsing_checkpoint, cuda)
image1 = 255 - tensor1.cpu().detach().numpy() * 255
image2 = 255 - tensor2.cpu().detach().numpy() * 255
shape = image1.shape
if len(shape) == 2:
image1 = image1[:, :, np.newaxis]
image2 = image2[:, :, np.newaxis]
img1 = ops.fill_gray(image1)
img2 = ops.fill_gray(image2)
img = ops.merge_4image(img1, img2, img3, img4)
cv2.imwrite(path, img)
p2 = os.path.join(path, "a-" + file)
al = align.face_features(p, p2)
ev = utils.parse_evaluate(al, args.parsing_checkpoint, cuda=cuda)
p = os.path.join(path, "b-" + file)
cv2.imwrite(p, ev)
ev = 255 - utils.img_edge(ev)
p = os.path.join(path, "c-" + file)
cv2.imwrite(p, ev)
elif args.phase == "dataset":
dataset = FaceDataset(args, "test")
dataset.pre_process(cuda)
elif args.phase == "preview":
log.info("preview picture")
path = "../export/regular/model.jpg"
img = cv2.imread(path)
img2 = utils.parse_evaluate(img, args.parsing_checkpoint, cuda)
img3 = utils.img_edge(img2)
img3_ = ops.fill_grey(img3)
img4 = align.face_features(path)
log.info("{0} {1} {2} {3}".format(img.shape, img2.shape, img3_.shape,
img4.shape))
ops.merge_4image(img, img2, img3_, img4, show=True)
elif args.phase == "evaluate":
log.info("evaluation mode start")
evl = Evaluate(args, cuda=cuda)
img = cv2.imread(args.eval_image).astype(np.float32)
x_ = evl.itr_train(img)
evl.output(x_, img)
else:
log.error("not known phase %s", args.phase)
