Getting Some Errrors

[FlowDownTheRiver]

Hi there.Using the same environment as Psfrgan for Face-Sparnet.
I am on Windows 10 64 bit.
I used this command line to test for a single image : python test_enhance_single_unalign.py --gpus 1 --model sparnethd --name SPARNetHD_V4_Attn2D --res_depth 10 --att_name spar --Gnorm 'in' --pretrain_model_path ./pretrain_models/SPARNetHD_V4_Attn2D_net_H-epoch10.pth --test_img_path ./test_images/test_hzgg.jpg --results_dir test_hzgg_results

and I am getting this in anaconda prompt line : (psfrgan) C:\Face-SPARNet>python test_enhance_single_unalign.py --gpus 1 --model sparnethd --name SPARNetHD_V4_Attn2D --res_depth 10 --att_name spar --Gnorm 'in' --pretrain_model_path ./pretrain_models/SPARNetHD_V4_Attn2D_net_H-epoch10.pth --test_img_path ./test_images/test_hzgg.jpg --results_dir test_hzgg_results
----------------- Options ---------------
Dnorm: none
Gnorm: 'in' [default: bn]
att_name: spar
batch_size: 32
bottleneck_size: 4
checkpoints_dir: ./check_points
data_device: cuda:0 [default: None]
dataroot: None
dataset_name: celeba
debug: False
device: cuda:0 [default: None]
epoch: latest
eval: False
gpu_ids: [0] [default: None]
gpus: 1
init_gain: 0.02
init_type: normal
input_nc: 3
isTrain: False [default: None]
load_iter: 0 [default: 0]
load_size: None
max_dataset_size: inf
model: sparnethd [default: test]
n_layers_D: 4
name: SPARNetHD_V4_Attn2D [default: experiment_name]
ndf: 64
ngf: 64
ntest: inf
num_D: 3
num_test: 50
num_threads: 8
output_nc: 3
phase: test
preprocess: none
pretrain_model_path: ./pretrain_models/SPARNetHD_V4_Attn2D_net_H-epoch10.pth [default: ]
res_depth: 10
results_dir: test_hzgg_results [default: ./results/]
save_as_dir:
seed: 123
serial_batches: False
suffix:
test_img_path: ./test_images/test_hzgg.jpg [default: ]
test_upscale: 1
verbose: False
----------------- End -------------------
======> Loading images, crop and align faces.
======> Saving aligned LQ faces to test_hzgg_results\LQ_faces
Traceback (most recent call last):
File "test_enhance_single_unalign.py", line 105, in
enhance_model = def_models(opt)
File "test_enhance_single_unalign.py", line 50, in def_models
model = create_model(opt)
File "C:\Face-SPARNet\models_init_.py", line 65, in create_model
instance = model(opt)
File "C:\Face-SPARNet\models\sparnethd_model.py", line 26, in init
res_depth=opt.res_depth, norm_type=opt.Gnorm, att_name=opt.att_name, bottleneck_size=opt.bottleneck_size)
File "C:\Face-SPARNet\models\sparnet.py", line 44, in init
self.encoder.append(ResidualBlock(cin, cout, scale='down', hg_depth=hg_depth, att_name=att_name, **nrargs))
File "C:\Face-SPARNet\models\blocks.py", line 119, in init
NormLayer(c_in, norm_type=self.norm_type),
File "C:\Face-SPARNet\models\blocks.py", line 30, in init
assert 1==0, 'Norm type {} not support.'.format(norm_type)
AssertionError: Norm type 'in' not support.

I am not sure what is wrong. I used git clone Face-Sparnet but I also tried the same thing in PSFRGAN folder as well.All pre-trained models are in place as your description.

[chaofengc]

The command looks fine. Possibly because the shell in windows parse the command in a different way. You may try to set the default Gnorm to 'in' in base_options.py and try again.

[FlowDownTheRiver]

I have set it from bg to in. Tried it. Still getting errors. I also tried deleting the --Gnorm from command line or the arguments related to it.Some lines are still complaining.

command line after changing Gnorm to in : python test_enhance_single_unalign.py --gpus 1 --model sparnethd --name SPARNetHD_V4_Attn2D --res_depth 10 --att_name spar --Gnorm 'in' --pretrain_model_path .\check_points\SPARNetHD_V4_Attn2D\latest_net_H.pth --test_img_path .\test_images\test_hzgg.jpg --results_dir test_hzgg_results

This is what I get as error : ----------------- End -------------------
======> Loading images, crop and align faces.
======> Saving aligned LQ faces to test_hzgg_results\LQ_faces
Traceback (most recent call last):
File "test_enhance_single_unalign.py", line 105, in
enhance_model = def_models(opt)
File "test_enhance_single_unalign.py", line 50, in def_models
model = create_model(opt)
File "C:\Face-SPARNet\models_init_.py", line 65, in create_model
instance = model(opt)
File "C:\Face-SPARNet\models\sparnethd_model.py", line 26, in init
res_depth=opt.res_depth, norm_type=opt.Gnorm, att_name=opt.att_name, bottleneck_size=opt.bottleneck_size)
File "C:\Face-SPARNet\models\sparnet.py", line 44, in init
self.encoder.append(ResidualBlock(cin, cout, scale='down', hg_depth=hg_depth, att_name=att_name, **nrargs))
File "C:\Face-SPARNet\models\blocks.py", line 119, in init
NormLayer(c_in, norm_type=self.norm_type),
File "C:\Face-SPARNet\models\blocks.py", line 30, in init
assert 1==0, 'Norm type {} not support.'.format(norm_type)
AssertionError: Norm type 'in' not support.

[FlowDownTheRiver]

I also get this :
(psfrgan) C:\Face-SPARNet>python test.py --gpus 1 --model sparnethd --name SPARNetHD_V4_Attn2D --res_depth 10 --att_name spar --Gnorm 'in' --load_size 512 --dataset_name single --dataroot test_dirs/CelebA-TestN/ --pretrain_model_path ./pretrain_models/SPARNetHD_V4_Attn2D_net_H-epoch10.pth --save_as_dir results_CelebA-TestN/SPARNetHD_V4_Attn2D/
----------------- Options ---------------
Dnorm: none
Gnorm: 'in' [default: in]
att_name: spar
batch_size: 32
bottleneck_size: 4
checkpoints_dir: ./check_points
data_device: cuda:0 [default: None]
dataroot: test_dirs/CelebA-TestN/ [default: None]
dataset_name: single [default: celeba]
debug: False
device: cuda:0 [default: None]
epoch: latest
eval: False
gpu_ids: [0] [default: None]
gpus: 1
init_gain: 0.02
init_type: normal
input_nc: 3
isTrain: False [default: None]
load_iter: 0 [default: 0]
load_size: 512 [default: None]
max_dataset_size: inf
model: sparnethd [default: test]
n_layers_D: 4
name: SPARNetHD_V4_Attn2D [default: experiment_name]
ndf: 64
ngf: 64
ntest: inf
num_D: 3
num_test: 50
num_threads: 8
output_nc: 3
phase: test
preprocess: none
pretrain_model_path: ./pretrain_models/SPARNetHD_V4_Attn2D_net_H-epoch10.pth [default: ]
res_depth: 10
results_dir: ./results/
save_as_dir: results_CelebA-TestN/SPARNetHD_V4_Attn2D/ [default: ]
seed: 123
serial_batches: False
suffix:
test_img_path:
test_upscale: 1
verbose: False
----------------- End -------------------
dataset [SingleDataset] was created
Traceback (most recent call last):
File "test.py", line 18, in
model = create_model(opt) # create a model given opt.model and other options
File "C:\Face-SPARNet\models_init_.py", line 65, in create_model
instance = model(opt)
File "C:\Face-SPARNet\models\sparnethd_model.py", line 26, in init
res_depth=opt.res_depth, norm_type=opt.Gnorm, att_name=opt.att_name, bottleneck_size=opt.bottleneck_size)
File "C:\Face-SPARNet\models\sparnet.py", line 44, in init
self.encoder.append(ResidualBlock(cin, cout, scale='down', hg_depth=hg_depth, att_name=att_name, **nrargs))
File "C:\Face-SPARNet\models\blocks.py", line 119, in init
NormLayer(c_in, norm_type=self.norm_type),
File "C:\Face-SPARNet\models\blocks.py", line 30, in init
assert 1==0, 'Norm type {} not support.'.format(norm_type)
AssertionError: Norm type 'in' not support.

[chaofengc]

If you set default Gnorm to in, you should remove the --Gnorm in the command.

From the error message "Norm type 'in' not support.", I guess the single quotes might be the problem. You may try to remove the quote. If this does not solve the problem, you may try to print the norm_type to debug the NormLayer class in blocks.py and see what happened.

BTW, the path separator may also cause error, because windowns use \ while Ubuntu use /.

[FlowDownTheRiver]

ok I finally got it working as you suggested. it is the quotes,I removed the --Gnorm and argument,launched the code and get it working like this :

(worked) python test.py --gpus 1 --model sparnethd --name SPARNetHD_V4_Attn2D --res_depth 10 --att_name spar --load_size 512 --dataset_name single --dataroot test_dirs/CelebA-TestN/ --pretrain_model_path ./pretrain_models/SPARNetHD_V4_Attn2D_net_H-epoch10.pth --save_as_dir results_CelebA-TestN/SPARNetHD_V4_Attn2D/

(worked) python test_enhance_single_unalign.py --gpus 1 --model sparnethd --name SPARNetHD_V4_Attn2D --res_depth 10 --att_name spar --pretrain_model_path .\pretrain_models\SPARNetHD_V4_Attn2D_net_H-epoch10.pth --test_img_path .\test_images\test_hzgg.jpg --results_dir test_hzgg_results

(didn't work) python test.py --gpus 1 --model sparnethd --name SPARNetHD_V4_Attn2D --res_depth 10 --att_name spar --Gnorm 'in' --load_size 512 --dataset_name single --dataroot test_dirs/CelebA-TestN/ --pretrain_model_path ./pretrain_models/SPARNetHD_V4_Attn2D_net_H-epoch10.pth --save_as_dir results_CelebA-TestN/SPARNetHD_V4_Attn2D/

[FlowDownTheRiver]

I have another question,is the Face-Sparnet repo also compatible with test_enhance_dir_align.py,test_enhance_dir_unalign.py in PSFRGAN repo,if I use these command lines and the py file together?

[chaofengc]

You will have to make some necessary modifications, because the inputs and network options of these two are different.

[justinjohn0306]

If you set default Gnorm to in, you should remove the --Gnorm in the command.

From the error message "Norm type 'in' not support.", I guess the single quotes might be the problem. You may try to remove the quote. If this does not solve the problem, you may try to print the norm_type to debug the NormLayer class in blocks.py and see what happened.

BTW, the path separator may also cause error, because windowns use \ while Ubuntu use /.

use --Gnorm in [for windows users]