train issue

[bye111]

训练无法生成pth文件，一直报路径错误，请问为什么
OSError: Failed to find fully trained model in 'outdir\celeba_preproc/model_d(cyclegan)_m(cyclegan)_d(basic)_g(vit-unet)_cyclegan_vit-unet-12_male2female_self-lsgan-paper-cycle_mid-256' for transfer: [Errno 2] No such file or directory: 'outdir\celeba_preproc/model_d(cyclegan)_m(cyclegan)_d(basic)_g(vit-unet)_cyclegan_vit-unet-12_male2female_self-lsgan-paper-cycle_mid-256\net_gen_ab.pth'

[usert5432]

Hi @bye111,

That is a strange error. Would it be possible for you to show the command you are running? I would help us to diagnose the issue better.

[bye111]

(Paint) E:\pycharm1\uvcgan2-main\uvcgan2-main>python scripts/anime2selfie/train_anime2selfie_translation.py
Starting training...
{
"batch_size": 1,
"data": {
"datasets": [
{
"dataset": {
"name": "cyclegan",
"domain": "a",
"path": "selfie2anime"
},
"shape": [
3,
256,
256
],
"transform_train": [
{
"name": "resize",
"size": 286
},
{
"name": "random-crop",
"size": 256
},
"random-flip-horizontal"
],
"transform_test": null
},
{
"dataset": {
"name": "cyclegan",
"domain": "b",
"path": "selfie2anime"
},
"shape": [
3,
256,
256
],
"transform_train": [
{
"name": "resize",
"size": 286
},
{
"name": "random-crop",
"size": 256
},
"random-flip-horizontal"
],
"transform_test": null
}
],
"merge_type": "unpaired",
"workers": 1
},
"epochs": 500,
"discriminator": {
"model": "basic",
"model_args": {
"shrink_output": false
},
"optimizer": {
"name": "Adam",
"lr": 0.0001,
"betas": [
0.5,
0.99
]
},
"weight_init": {
"name": "normal",
"init_gain": 0.02
},
"spectr_norm": true
},
"generator": {
"model": "vit-modnet",
"model_args": {
"features": 384,
"n_heads": 6,
"n_blocks": 12,
"ffn_features": 1536,
"embed_features": 384,
"activ": "gelu",
"norm": "layer",
"modnet_features_list": [
48,
96,
192,
384
],
"modnet_activ": "leakyrelu",
"modnet_norm": null,
"modnet_downsample": "conv",
"modnet_upsample": "upsample-conv",
"modnet_rezero": false,
"modnet_demod": true,
"rezero": true,
"activ_output": "sigmoid",
"style_rezero": true,
"style_bias": true,
"n_ext": 1
},
"optimizer": {
"name": "Adam",
"lr": 5e-05,
"betas": [
0.5,
0.99
]
},
"weight_init": {
"name": "normal",
"init_gain": 0.02
},
"spectr_norm": false
},
"model": "uvcgan2",
"model_args": {
"lambda_a": 10.0,
"lambda_b": 10.0,
"lambda_idt": 0.5,
"avg_momentum": 0.9999,
"head_queue_size": 3,
"head_config": {
"name": "batch-norm-2d",
"input_features": 512,
"output_features": 1,
"activ": "leakyrelu"
}
},
"loss": "lsgan",
"gradient_penalty": {
"center": 0,
"lambda_gp": 0.01,
"mix_type": "real-fake",
"reduction": "mean"
},
"seed": 0,
"scheduler": null,
"steps_per_epoch": 2000,
"transfer": {
"base_model": "anime2selfie/anime2selfie/model_m(uvcgan2)_d(basic)_g(vit-modnet)_pretrain-uvcgan2",
"transfer_map": {
"gen_ab": "encoder",
"gen_ba": "encoder"
},
"strict": true,
"allow_partial": false,
"fuzzy": null
}
}
D:\Anaconda\envs\Paint\lib\site-packages\torch\functional.py:504: UserWarning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument. (Triggered internally at C:\cb\pytorch_1000000000000\work\aten\src\ATen\native\TensorShape.cpp:3484.)
return _VF.meshgrid(tensors, **kwargs) # type: ignore[attr-defined]
[2025-05-14 14:40:22,853] [uvcgan2.base]: DEBUG Initializnig network with normal
[2025-05-14 14:40:22,991] [uvcgan2.base]: DEBUG Initializnig network with normal
[2025-05-14 14:40:23,122] [uvcgan2.base]: DEBUG Initializnig network with normal
[2025-05-14 14:40:23,264] [uvcgan2.base]: DEBUG Initializnig network with normal
[2025-05-14 14:40:23,291] [uvcgan2.base]: DEBUG Initializnig network with normal
[2025-05-14 14:40:23,353] [uvcgan2.base]: DEBUG Initializnig network with normal
[2025-05-14 14:40:23,381] [uvcgan2.train]: INFO Initiating parameter transfer : '{'base_model': 'anime2selfie/anime2selfie/model_m(uvcgan2)_d(basic)_g(vit-modnet)_pretrain-uvcgan2', 'transfer_map': {'gen_ab': 'encoder', 'gen_ba': 'encoder'}, 'strict': True, 'allow_partial': False, 'fuzzy': None}'
[2025-05-14 14:40:23,537] [uvcgan2.base]: DEBUG Initializnig network with normal
[2025-05-14 14:40:23,676] [uvcgan2.base]: DEBUG Initializnig network with normal
[2025-05-14 14:40:23,804] [uvcgan2.base]: DEBUG Initializnig network with normal
[2025-05-14 14:40:23,938] [uvcgan2.base]: DEBUG Initializnig network with normal
[2025-05-14 14:40:23,961] [uvcgan2.base]: DEBUG Initializnig network with normal
[2025-05-14 14:40:23,986] [uvcgan2.base]: DEBUG Initializnig network with normal
[2025-05-14 14:40:24,000] [uvcgan2.cgan]: DEBUG Loading model from epoch None
Traceback (most recent call last):
File "e:\pycharm1\uvcgan2-main\uvcgan2-main\uvcgan2\train\transfer.py", line 14, in load_base_model
model.load(epoch = None)
File "e:\pycharm1\uvcgan2-main\uvcgan2-main\uvcgan2\cgan\model_base.py", line 110, in load
load(self.models, self.savedir, PREFIX_MODEL, epoch, self.device)
File "e:\pycharm1\uvcgan2-main\uvcgan2-main\uvcgan2\cgan\checkpoint.py", line 71, in load
torch.load(load_path, map_location = device)
File "D:\Anaconda\envs\Paint\lib\site-packages\torch\serialization.py", line 791, in load
with _open_file_like(f, 'rb') as opened_file:
File "D:\Anaconda\envs\Paint\lib\site-packages\torch\serialization.py", line 271, in _open_file_like
return _open_file(name_or_buffer, mode)
File "D:\Anaconda\envs\Paint\lib\site-packages\torch\serialization.py", line 252, in init
super().init(open(name, mode))
FileNotFoundError: [Errno 2] No such file or directory: 'outdir\anime2selfie/anime2selfie/model_m(uvcgan2)_d(basic)_g(vit-modnet)_pretrain-uvcgan2\net_gen_ab.pth'

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
File "E:\pycharm1\uvcgan2-main\uvcgan2-main\scripts\anime2selfie\train_anime2selfie_translation.py", line 142, in
train(args_dict)
File "e:\pycharm1\uvcgan2-main\uvcgan2-main\uvcgan2\train\train.py", line 69, in train
transfer(model, args.transfer)
File "e:\pycharm1\uvcgan2-main\uvcgan2-main\uvcgan2\train\transfer.py", line 163, in transfer
base_model = get_base_model(transfer_config, model.device)
File "e:\pycharm1\uvcgan2-main\uvcgan2-main\uvcgan2\train\transfer.py", line 48, in get_base_model
load_base_model(model, transfer_config)
File "e:\pycharm1\uvcgan2-main\uvcgan2-main\uvcgan2\train\transfer.py", line 19, in load_base_model
raise IOError(
OSError: Failed to find fully trained model in 'outdir\anime2selfie/anime2selfie/model_m(uvcgan2)_d(basic)_g(vit-modnet)_pretrain-uvcgan2' for transfer: [Errno 2] No such file or directory: 'outdir\anime2selfie/anime2selfie/model_m(uvcgan2)_d(basic)_g(vit-modnet)_pretrain-uvcgan2\net_gen_ab.pth'

[bye111]

Hi @bye111,  你好

That is a strange error. Would it be possible for you to show the command you are running? I would help us to diagnose the issue better.这是一个奇怪的错误。您能否显示正在运行的命令？我会帮助我们更好地诊断问题。

what i'm confused is how to generate net_gen_ab.pth . And the command im running is train_anime2selfie_translation.
Does the net_gen_ab.pth generated by pretrain model ? Thanks for your reply

[usert5432]

Hi @bye111,

Thank you for sharing the additional details.

I think, the problem may be related to the fact that the training script tries to load a pre-trained image inpainting model, but cannot find it:

"base_model": "anime2selfie/anime2selfie/model_m(uvcgan2)_d(basic)_g(vit-modnet)_pretrain-uvcgan2",

what i'm confused is how to generate net_gen_ab.pth . And the command im running is train_anime2selfie_translation.
Does the net_gen_ab.pth generated by pretrain model ? Thanks for your reply

I think, this mostly depends on your goals. Would you prefer to train the transfer model (net_gen_ab.pth) yourself, or use a model that we shared with the paper? Likewise, if you prefer to train the model yourself, would you prefer to pre-train image inpainting model first, or use the one that we shared with the paper?

Overall, the easiest way to fix the error is to set the entire transfer block from:

 "transfer": {
"base_model": "anime2selfie/anime2selfie/model_m(uvcgan2)_d(basic)_g(vit-modnet)_pretrain-uvcgan2",
"transfer_map": {
"gen_ab": "encoder",
"gen_ba": "encoder"
},

to simply

 "transfer":  None,

That will completely disable the use of pre-trained inpainting models and the error should go away. The training script should run and create net_gen_ab.pth. However, the final translation performance may be suboptimal.

Please let me know if you would like to use the pre-trained image inpainting model, and I would be happy to walk you though the required steps.

[bye111]

Hi @bye111,  你好

Thank you for sharing the additional details.感谢您分享更多详细信息。

I think, the problem may be related to the fact that the training script tries to load a pre-trained image inpainting model, but cannot find it:我认为，问题可能与训练脚本尝试加载预训练图像修复模型，但找不到它有关：

"base_model": "anime2selfie/anime2selfie/model_m(uvcgan2)_d(basic)_g(vit-modnet)_pretrain-uvcgan2",

what i'm confused is how to generate net_gen_ab.pth . And the command im running is train_anime2selfie_translation.我困惑的是如何生成 net_gen_ab.pth 。我运行的命令是 train_anime2selfie_translation。
Does the net_gen_ab.pth generated by pretrain model ? Thanks for your reply预训练模型生成的 net_gen_ab.pth 是否？感谢您的回复

I think, this mostly depends on your goals. Would you prefer to train the transfer model (net_gen_ab.pth) yourself, or use a model that we shared with the paper? Likewise, if you prefer to train the model yourself, would you prefer to pre-train image inpainting model first, or use the one that we shared with the paper?我认为，这主要取决于你的目标。您更愿意自己训练迁移模型 （net_gen_ab.pth），还是使用我们与论文共享的模型？同样，如果你更喜欢自己训练模型，你更愿意先预训练图像修复模型，还是使用我们与论文共享的模型？

Overall, the easiest way to fix the error is to set the entire transfer block from:总的来说，修复错误的最简单方法是从以下位置设置整个 transfer 块：

"transfer": {
"base_model": "anime2selfie/anime2selfie/model_m(uvcgan2)_d(basic)_g(vit-modnet)_pretrain-uvcgan2",
"transfer_map": {
"gen_ab": "encoder",
"gen_ba": "encoder"
},
to simply  改为 simple

"transfer": None,
That will completely disable the use of pre-trained inpainting models and the error should go away. The training script should run and create net_gen_ab.pth. However, the final translation performance may be suboptimal.这将完全禁用预训练的修复模型，错误应该会消失。训练脚本应运行并创建 net_gen_ab.pth。但是，最终的翻译性能可能不是最佳的。

Please let me know if you would like to use the pre-trained image inpainting model, and I would be happy to walk you though the required steps.如果您想使用预训练的图像修复模型，请告诉我，我很乐意指导您完成所需的步骤。

I prefer to train the transfer model (net_gen_ab.pth) and translation model myself. Does it means that we could use '--no-pretrain' command to train my own datasets ?
I‘d like to use the pre-trained image inpainting and translating model ,does them trained on special datasets such as anime2selfie? How could trained on my own datasets .Looking forward to your reply

[usert5432]

Hi @bye111,

Sorry, I am not entirely sure how to interpret your requirements, so I am not sure if I can give an actionable reply. Incidentally, would you find it more convenient to communicate in Chinese? I can ask my colleague help with the conversation.

[bye111]

Hi @bye111,  你好

Sorry, I am not entirely sure how to interpret your requirements, so I am not sure if I can give an actionable reply. Incidentally, would you find it more convenient to communicate in Chinese? I can ask my colleague help with the conversation.抱歉，我不完全确定如何解释您的要求，因此我不确定我是否可以给出可行的答复。顺便说一句，你会觉得用中文交流更方便吗？我可以请我的同事帮忙进行对话。

Thanks for your reply. It's ok to use English , I use command '--no-pretrain' to solve my problem