Finetuning provided model about gfpgan HOT 6 CLOSED

@icoric4
The training setting is similar, but

1. we remove the component loss
2. we use more data and data with higher quality
3. the training network structure is not the same as the inference network structure

from gfpgan.

@icoric4 Finetuning the https://github.com/TencentARC/GFPGAN/releases/download/v0.2.0/GFPGANCleanv1-NoCE-C2.pth has not been supported yet. As this model is converted from another model (which is not released.)

I will add the model, but it may take much time ~

from gfpgan.

Thanks!

from gfpgan.

Hello, I have another question.

When training from scratch a model like this one https://github.com/TencentARC/GFPGAN/releases/download/v0.2.0/GFPGANCleanv1-NoCE-C2.pth,

could you share the training settings you have used to train it, similar to this?

# training settings
train:
  optim_g:
    type: Adam
    lr: !!float 2e-3
  optim_d:
    type: Adam
    lr: !!float 2e-3
  optim_component:
    type: Adam
    lr: !!float 2e-3

  scheduler:
    type: MultiStepLR
    milestones: [600000, 700000]
    gamma: 0.5

  total_iter: 800000
  warmup_iter: -1  # no warm up

  # losses
  # pixel loss
  pixel_opt:
    type: L1Loss
    loss_weight: !!float 1e-1
    reduction: mean
  # L1 loss used in pyramid loss, component style loss and identity loss
  L1_opt:
    type: L1Loss
    loss_weight: 1
    reduction: mean

  # image pyramid loss
  pyramid_loss_weight: 1
  remove_pyramid_loss: 50000
  # perceptual loss (content and style losses)
  perceptual_opt:
    type: PerceptualLoss
    layer_weights:
      # before relu
      'conv1_2': 0.1
      'conv2_2': 0.1
      'conv3_4': 1
      'conv4_4': 1
      'conv5_4': 1
    vgg_type: vgg19
    use_input_norm: true
    perceptual_weight: !!float 1
    style_weight: 50
    range_norm: true
    criterion: l1
  # gan loss
  gan_opt:
    type: GANLoss
    gan_type: wgan_softplus
    loss_weight: !!float 1e-1
  # r1 regularization for discriminator
  r1_reg_weight: 10
  # facial component loss
  gan_component_opt:
    type: GANLoss
    gan_type: vanilla
    real_label_val: 1.0
    fake_label_val: 0.0
    loss_weight: !!float 1
  comp_style_weight: 200
  # identity loss
  identity_weight: 10

  net_d_iters: 1
  net_d_init_iters: 0
  net_d_reg_every: 16

from gfpgan.

你好，如果我想训练 type: GFPGANv1Clean，该怎样获取正确的pre-trained stylegan2 model呢？

如此设置时 decoder_load_path: experiments/pretrained_models/StyleGAN2_512_Cmul1_FFHQ_B12G4_scratch_800k.pth
会报如下错误：

RuntimeError: Error(s) in loading state_dict for StyleGAN2GeneratorCSFT:
Missing key(s) in state_dict: "style_mlp.9.weight", "style_mlp.9.bias", "style_mlp.11.weight", "style_mlp.11.bias", "style_mlp.13.weight", "style_mlp.13.bias", "style_mlp.15.weight", "style_mlp.15.bias", "style_conv1.bias", "style_convs.0.bias", "style_convs.1.bias", "style_convs.2.bias", "style_convs.3.bias", "style_convs.4.bias", "style_convs.5.bias", "style_convs.6.bias", "style_convs.7.bias", "style_convs.8.bias", "style_convs.9.bias", "style_convs.10.bias", "style_convs.11.bias", "style_convs.12.bias", "style_convs.13.bias".
Unexpected key(s) in state_dict: "style_mlp.2.weight", "style_mlp.2.bias", "style_mlp.4.weight", "style_mlp.4.bias", "style_mlp.6.weight", "style_mlp.6.bias", "style_mlp.8.weight", "style_mlp.8.bias", "style_conv1.activate.bias", "style_convs.0.activate.bias", "style_convs.1.activate.bias", "style_convs.2.activate.bias", "style_convs.3.activate.bias", "style_convs.4.activate.bias", "style_convs.5.activate.bias", "style_convs.6.activate.bias", "style_convs.7.activate.bias", "style_convs.8.activate.bias", "style_convs.9.activate.bias", "style_convs.10.activate.bias", "style_convs.11.activate.bias", "style_convs.12.activate.bias", "style_convs.13.activate.bias".

from gfpgan.

你好，如果我想训练 type: GFPGANv1Clean，该怎样获取正确的pre-trained stylegan2 model呢？

如此设置时 decoder_load_path: experiments/pretrained_models/StyleGAN2_512_Cmul1_FFHQ_B12G4_scratch_800k.pth 会报如下错误：

RuntimeError: Error(s) in loading state_dict for StyleGAN2GeneratorCSFT: Missing key(s) in state_dict: "style_mlp.9.weight", "style_mlp.9.bias", "style_mlp.11.weight", "style_mlp.11.bias", "style_mlp.13.weight", "style_mlp.13.bias", "style_mlp.15.weight", "style_mlp.15.bias", "style_conv1.bias", "style_convs.0.bias", "style_convs.1.bias", "style_convs.2.bias", "style_convs.3.bias", "style_convs.4.bias", "style_convs.5.bias", "style_convs.6.bias", "style_convs.7.bias", "style_convs.8.bias", "style_convs.9.bias", "style_convs.10.bias", "style_convs.11.bias", "style_convs.12.bias", "style_convs.13.bias". Unexpected key(s) in state_dict: "style_mlp.2.weight", "style_mlp.2.bias", "style_mlp.4.weight", "style_mlp.4.bias", "style_mlp.6.weight", "style_mlp.6.bias", "style_mlp.8.weight", "style_mlp.8.bias", "style_conv1.activate.bias", "style_convs.0.activate.bias", "style_convs.1.activate.bias", "style_convs.2.activate.bias", "style_convs.3.activate.bias", "style_convs.4.activate.bias", "style_convs.5.activate.bias", "style_convs.6.activate.bias", "style_convs.7.activate.bias", "style_convs.8.activate.bias", "style_convs.9.activate.bias", "style_convs.10.activate.bias", "style_convs.11.activate.bias", "style_convs.12.activate.bias", "style_convs.13.activate.bias".

@timfu248 有解决吗兄弟，目前是不是只能微调GFPGANv1.pth呀

from gfpgan.