Pytorch-cyclegan-and-pix2pix: AttributeError: 'Sequential' object has no attribute 'model'

[wangxiaodiu]

I have some satellite(aerial) images, all are 256x256, and I put them in * ~/temp/*, I want to translate them into map, so I downloaded the pretrained model and use the following command to test:

CUDA_VISIBLE_DEVICES=1 python test.py --dataroot ~/temp/ --dataset_mode single --resize_or_crop none --which_direction AtoB --model test --name sat2map_pretrained

But I have got an AttributeError from python , here is the output:

----------------- Options ---------------
aspect_ratio: 1.0
batchSize: 1
checkpoints_dir: ./checkpoints
dataroot: /home/niu/temp/ [default: None]
dataset_mode: single
display_env: main
display_id: 1
display_port: 8097
display_server: http://localhost
display_winsize: 256
fineSize: 256
gpu_ids: 0
how_many: 50
init_gain: 0.02
init_type: normal
input_nc: 3
isTrain: False [default: None]
loadSize: 256
max_dataset_size: inf
model: test

▽
model_suffix:
nThreads: 4
n_layers_D: 3
name: sat2map_pretrained [default: experiment_name]
ndf: 64
ngf: 64
no_dropout: True
no_flip: False
norm: instance
ntest: inf
output_nc: 3
phase: test
resize_or_crop: none [default: resize_and_crop]
results_dir: ./results/
serial_batches: False
suffix:
verbose: False
which_direction: AtoB
which_epoch: latest
which_model_netD: basic
which_model_netG: resnet_9blocks
----------------- End -------------------
dataset [SingleImageDataset] was created
initialize network with normal
model [TestModel] was created
loading the model from ./checkpoints/sat2map_pretrained/latest_net_G.pth
Traceback (most recent call last):
File "test.py", line 19, in
model.setup(opt)
File "/home/niu/src/github/pytorch-CycleGAN-and-pix2pix/models/base_model.py", line 43, in setup
self.load_networks(opt.which_epoch)
File "/home/niu/src/github/pytorch-CycleGAN-and-pix2pix/models/base_model.py", line 135, in load_networks
self.__patch_instance_norm_state_dict(state_dict, net, key.split('.'))
File "/home/niu/src/github/pytorch-CycleGAN-and-pix2pix/models/base_model.py", line 115, in __patch_instance_norm_state_dict
self.__patch_instance_norm_state_dict(state_dict, getattr(module, key), keys, i + 1)
File "/home/niu/src/github/pytorch-CycleGAN-and-pix2pix/models/base_model.py", line 115, in __patch_instance_norm_state_dict
self.__patch_instance_norm_state_dict(state_dict, getattr(module, key), keys, i + 1)
File "/home/niu/anaconda3/envs/pix2pix/lib/python3.7/site-packages/torch/nn/modules/module.py", line 518, in __getattr__
type(self).__name__, name))
AttributeError: 'Sequential' object has no attribute 'model'

I have looked into similat issues and googled for an hour but got nothing.
Could someone help me with this? Thank you!

[taesungp]

hmmmm..... are you using the latest code and the model? Could you git pull the latest code, and also download the latest pretrained model, and try it again? I think I vaguely remember our code had this kind of bug a while ago...

[Kanata-Bifang]

I wonder if you've solved this issue.
I met the same problem when testing a cyclegan model, but it didn't happen in my test of pix2pix model before.

[junyanz]

Are you using the latest code? Which PyTorch version are you using?

[Kanata-Bifang]

yeah. I'm using the latest code with PyTorch 0.4.0

[jiemojiemo]

you need to set --netG option if you are not using the default netG

[Kanata-Bifang]

i found the problem was caused by the incorrect input of --model. thank you for your reply

[shimiaoli]

Hi, I have encountered the same error. I found this error occurred due to running a pix2pix model uisng --model test . I have changed line 14 in models/test_model.py from parser = CycleGANModel.modify_commandline_options(parser, is_train=False) to parser = Pix2PixModel.modify_commandline_options(parser, is_train = False) . Also I added from .pix2pix_model import Pix2PixModel at the beginning of the file. Then it runs well. Hope this helps.

[junyanz]

For pix2pix, you just need to use --model pix2pix for testing. See the instruction. --model test is useful when you want to apply cyclegan in one direction.

[shimiaoli]

Actually I want to use --dataset_mode single then it seems I must use --model test to go with it

[junyanz]

I see. You can also specify --netG unet --norm batch for --model test.

[taozhuang123]

please tell me , i am beginner,i meet same question , i just change the netG, what can i do? thank you every much.

[yangzhikai]

@ taozhuang123 hi ，I got the trick， IN 'test_option.py' Line20 ,the model should set as: model='pix2pix'.....hhha~~~~

[guotong1988]

edit to
parser.set_defaults(model='pix2pix')
in
test_options.py

[wuzihaoo]

you need to set --netG option if you are not using the default netG

This works for me. Keep --netG the same as training

[theidealist2]

@ taozhuang123 hi ，I got the trick， IN 'test_option.py' Line20 ,the model should set as: model='pix2pix'.....hhha~~~~

you are right! thanks!