def get_test_result(a_jpg_path, b_jpg_path, model, opt):
### 1. 还要把val几个文件的JSON信息也储存到另外的文件，这样比直接读大JSON文件快。
a_parsing_path = a_jpg_path.replace('.jpg', '.png').replace('img/', 'img_parsing_all/')
#b_parsing_path = b_jpg_path.replace('.jpg', '.png').replace('img/', 'img_parsing_all/')
a_json_path = a_jpg_path.replace('.jpg', '_keypoints.json').replace('img/', 'img_keypoint_json/')
b_json_path = b_jpg_path.replace('.jpg', '_keypoints.json').replace('img/', 'img_keypoint_json/')
src = a_jpg_path.replace('.jpg', '_vis.png').replace('img/', 'img_parsing_all/').split(os.sep)
dst = b_jpg_path.replace('.jpg', '_vis.png').replace('img/', 'img_parsing_all/').split(os.sep)
theta_pair_key = src[-2] + '_' + src[-1] + "=" + dst[-2] + '_' + dst[-1]
# "id_00000006_01_1_front_TO_id_00000006_01_2_side__fake_b_parsing.png"
b_parsing_label_filename = theta_pair_key.replace('=', '_TO_')
b_parsing_label_filename = b_parsing_label_filename.replace('_vis.png', '') + '__fake_b_parsing.png'
b_parsing_path = os.path.join(opt.joint_test_data_dir, b_parsing_label_filename)
# a_parsing_tensor = get_parsing_label_tensor(a_parsing_path, self.opt)
b_parsing_tensor = get_parsing_label_tensor(b_parsing_path, opt)
# b_parsing_tensor = get_parsing_tensor(b_parsing_path, opt)
b_label_tensor, b_label_show_tensor = get_label_tensor(b_json_path, b_jpg_path, opt)
a_image_tensor = get_image_tensor(a_jpg_path, opt) # 这里的parsingRGB应该用20块好。本来就没采用10块计算。
b_image_tensor = get_image_tensor(b_jpg_path, opt)
# theta_aff_tensor, theta_tps_tensor, theta_aff_tps_tensor = get_thetas_tensor(theta_json_data, theta_pair_key) ### [1,18] --> 1*256*256
theta_aff_tensor, theta_aff_tps_tensor = get_thetas_affgrid_tensor(data_loader.dataset.affTnf, data_loader.dataset.tpsTnf, data_loader.dataset.theta_json_data, theta_pair_key)
input_tensor = torch.cat([a_image_tensor, b_image_tensor, b_label_tensor, b_parsing_tensor, theta_aff_tensor, theta_aff_tps_tensor], dim=0)
input_var = Variable(input_tensor[None, :, :, :].type(torch.cuda.FloatTensor))
if opt.isTrain:
fake_b = model.module.inference(input_var)
else:
fake_b = model.inference(input_var)
return fake_b, a_image_tensor, b_image_tensor, b_label_show_tensor, b_parsing_tensor
def main():
web_dir = os.path.join(
opt.results_dir, opt.name,
'%s_%s_%s' % (opt.phase, opt.which_epoch, "I_and_II_pose_seq"))
webpage = html.HTML(
web_dir, 'Experiment = %s, Phase = %s, Epoch = %s' %
(opt.name, opt.phase, opt.which_epoch))
for i, data in enumerate(dataset):
if i >= opt.how_many:
break
a_image_tensor = data['a_image_tensor'] # 3
b_image_tensor = data['b_image_tensor'] # 3
# b_label_tensor = data['b_label_tensor'] # 18
a_parsing_tensor = data['a_parsing_tensor'] # 1
# b_label_show_tensor = data['b_label_show_tensor']
a_jpg_path = data['a_jpg_path']
b_jpg_path = data['b_jpg_path']
a_parsing_rgb_tensor = parsingim_2_tensor(
a_parsing_tensor[0],
opt=opt,
parsing_label_nc=opt.parsing_label_nc)
# show_image_tensor_1 = torch.cat((a_image_tensor, a_parsing_rgb_tensor), dim=3)
show_image_tensor_1 = a_image_tensor
test_list = []
b_json_path_list = get_pose_seq_list(opt)
for j in range(len(b_json_path_list)):
b_json_path = b_json_path_list[j]
b_label_tensor, b_label_show_tensor = get_label_tensor(
b_json_path, b_jpg_path[0], opt)
b_label_tensor = b_label_tensor.unsqueeze_(0)
b_label_show_tensor = b_label_show_tensor.unsqueeze_(0)
fake_b, fake_b_parsing_label_tensor = generate_fake_B(
a_image_tensor, b_image_tensor, b_label_tensor,
a_parsing_tensor)
b_parsing_rgb_tensor = parsingim_2_tensor(
fake_b_parsing_label_tensor[0],
opt=opt,
parsing_label_nc=opt.parsing_label_nc)
# show_image_tensor_1 = torch.cat((show_image_tensor_1, b_label_show_tensor, b_parsing_rgb_tensor, fake_b.data[0:1, :, :, :].cpu()), dim=3)
show_image_tensor_1 = torch.cat(
(show_image_tensor_1, fake_b.data[0:1, :, :, :].cpu()), dim=3)
test_list.append(
('fake_image_seq', util.tensor2im(show_image_tensor_1[0])))
### save image
visuals = OrderedDict(test_list)
visualizer.save_images(webpage, visuals, a_jpg_path[0], b_jpg_path[0])
if i % 1 == 0:
print('[%s]process image... %s' % (i, a_jpg_path[0]))
webpage.save()
image_dir = webpage.get_image_dir()
print image_dir
def get_test_result(a_jpg_path, b_jpg_path, model, opt):
### 1. 还要把val几个文件的JSON信息也储存到另外的文件，这样比直接读大JSON文件快。
a_parsing_path = a_jpg_path.replace('.jpg', '.png').replace(
'img/', 'img_parsing_all/')
# b_parsing_path = b_jpg_path.replace('.jpg', '.png').replace('img/', 'img_parsing_all/')
# a_json_path = a_jpg_path.replace('.jpg', '_keypoints.json').replace('img/', 'img_keypoint_json/')
b_json_path = b_jpg_path.replace('.jpg', '_keypoints.json').replace(
'img/', 'img_keypoint_json/')
src = a_jpg_path.replace('.jpg', '_vis.png').replace(
'img/', 'img_parsing_all/').split(os.sep)
dst = b_jpg_path.replace('.jpg', '_vis.png').replace(
'img/', 'img_parsing_all/').split(os.sep)
theta_pair_key = src[-2] + '_' + src[-1] + "=" + dst[-2] + '_' + dst[-1]
b_parsing_label_filename = theta_pair_key.replace('=', '_TO_')
b_parsing_label_filename = b_parsing_label_filename.replace(
'_vis.png', '') + '__fake_b_parsing.png'
b_parsing_path = os.path.join(opt.joint_test_data_dir,
b_parsing_label_filename)
a_parsing_tensor = get_parsing_label_tensor(a_parsing_path, opt)
b_parsing_tensor = get_parsing_label_tensor(b_parsing_path, opt)
b_label_tensor, b_label_show_tensor = get_label_tensor(
b_json_path, b_jpg_path, opt)
a_image_tensor = get_image_tensor(a_jpg_path, opt)
b_image_tensor = get_image_tensor(b_jpg_path, opt)
a_parsing_rgb_tensor = parsingim_2_tensor(
a_parsing_tensor, opt=opt, parsing_label_nc=opt.parsing_label_nc)
b_parsing_rgb_tensor = parsingim_2_tensor(
b_parsing_tensor, opt=opt, parsing_label_nc=opt.parsing_label_nc)
theta_aff_tensor, theta_tps_tensor, theta_aff_tps_tensor = get_thetas_affgrid_tensor(
data_loader.dataset.affTnf, data_loader.dataset.tpsTnf,
data_loader.dataset.theta_json_data, theta_pair_key)
input_tensor = torch.cat([a_image_tensor, b_image_tensor, b_label_tensor, a_parsing_tensor, b_parsing_tensor, \
theta_aff_tensor, theta_tps_tensor, theta_aff_tps_tensor], dim=0)
input_var = Variable(input_tensor[None, :, :, :].type(
torch.cuda.FloatTensor))
model.eval()
if opt.isTrain:
fake_b = model.module.inference(input_var)
else:
fake_b = model.inference(input_var)
return fake_b, a_image_tensor, b_image_tensor, b_label_show_tensor
def get_test_result(a_jpg_path, b_jpg_path, model, opt):
# a_json_path = a_jpg_path.replace('.jpg', '_keypoints.json').replace('img/', 'img_keypoint_json/')
b_json_path = b_jpg_path.replace('.jpg', '_keypoints.json').replace(
'img/', 'img_keypoint_json/')
b_label_tensor, b_label_show_tensor = get_label_tensor(
b_json_path, b_jpg_path, opt)
a_image_tensor = get_image_tensor(a_jpg_path, opt)
b_image_tensor = get_image_tensor(b_jpg_path, opt)
input_tensor = torch.cat([a_image_tensor, b_image_tensor, b_label_tensor],
dim=0)
input_var = Variable(input_tensor[None, :, :, :].type(
torch.cuda.FloatTensor))
model.eval()
if opt.isTrain:
fake_b = model.module.inference(input_var)
else:
fake_b = model.inference(input_var)
return fake_b, a_image_tensor, b_image_tensor, b_label_show_tensor
def get_test_result(a_jpg_path, b_jpg_path, model, opt):
if 20 == opt.parsing_label_nc:
a_parsing_path = a_jpg_path.replace('.jpg', '.png').replace(
'img/', 'img_parsing_all/')
b_parsing_path = b_jpg_path.replace('.jpg', '.png').replace(
'img/', 'img_parsing_all/')
elif 10 == opt.parsing_label_nc:
a_parsing_path = a_jpg_path.replace('.jpg', '.png').replace(
'img/', 'img_parsing_all_10channel/')
b_parsing_path = b_jpg_path.replace('.jpg', '.png').replace(
'img/', 'img_parsing_all_10channel/')
b_json_path = b_jpg_path.replace('.jpg', '_keypoints.json').replace(
'img/', 'img_keypoint_json/')
# a_parsing_tensor = get_parsing_tensor(a_parsing_path)
# b_parsing_tensor = get_parsing_tensor(b_parsing_path)
a_parsing_tensor = get_parsing_label_tensor(a_parsing_path, opt)
b_parsing_tensor = get_parsing_label_tensor(b_parsing_path, opt)
b_label_tensor, b_label_show_tensor, _ = get_label_tensor(
b_json_path, b_jpg_path, opt)
input_dict = {
'b_label_tensor': b_label_tensor, \
'a_parsing_tensor': a_parsing_tensor, \
'b_parsing_tensor': b_parsing_tensor, \
'b_label_show_tensor': b_label_show_tensor}
# input_tensors = torch.cat((a_parsing_tensor, b_parsing_tensor, b_label_tensor), dim=0)
# input_var = Variable(input_tensors[None, :, :, :].type(torch.cuda.FloatTensor)) ##torch.FloatTensor of size (1,34,256,256)
# pdb.set_trace()
model.eval()
if opt.isTrain:
fake_b_parsing = model.module.inference(input_dict)
else:
fake_b_parsing = model.inference(input_var)
return a_parsing_tensor, b_parsing_tensor, fake_b_parsing, b_label_show_tensor