GAN模型PNNX转化问题

[zhu-zhaofei]

error log | 日志或报错信息 | ログ

描述：pnnx转换，模型不能生成ncnn.parm和 ncnn.bin文件
日志：
pnnxparam = E://opt//ncnn//tools//pnnx//build_debug//install//bin//GFPGANCleanv1-NoCE-C2.pnnx.param
pnnxbin = E://opt//ncnn//tools//pnnx//build_debug//install//bin//GFPGANCleanv1-NoCE-C2.pnnx.bin
pnnxpy = E://opt//ncnn//tools//pnnx//build_debug//install//bin//GFPGANCleanv1-NoCE-C2.pnnx.py
optlevel = 2
device = cpu
inputshape = [1,3,512,512]f32
inputshape2 =
customop =
moduleop =
############# pass_level0
inline module = gfpgan.archs.gfpganv1_clean_arch.ResBlock
inline module = gfpgan.archs.gfpganv1_clean_arch.StyleGAN2GeneratorCSFT
inline module = gfpgan.archs.stylegan2_clean_arch.ConstantInput
inline module = gfpgan.archs.stylegan2_clean_arch.ModulatedConv2d
inline module = gfpgan.archs.stylegan2_clean_arch.StyleConv
inline module = gfpgan.archs.stylegan2_clean_arch.ToRGB
############# pass_level1
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
unknown Parameter value kind prim::Constant
no attribute value
no attribute value
no attribute value
no attribute value
unknown Parameter value kind prim::Constant
no attribute value
no attribute value
no attribute value
no attribute value
unknown Parameter value kind prim::Constant
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
unknown Parameter value kind prim::Constant
no attribute value
no attribute value
no attribute value
no attribute value
unknown Parameter value kind prim::Constant
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
unknown Parameter value kind prim::Constant
no attribute value
no attribute value
no attribute value
no attribute value
unknown Parameter value kind prim::Constant
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
unknown Parameter value kind prim::Constant
no attribute value
no attribute value
no attribute value
no attribute value
unknown Parameter value kind prim::Constant
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
unknown Parameter value kind prim::Constant
no attribute value
no attribute value
no attribute value
no attribute value
unknown Parameter value kind prim::Constant
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
unknown Parameter value kind prim::Constant
no attribute value
no attribute value
no attribute value
no attribute value
unknown Parameter value kind prim::Constant
no attribute value
no attribute value
no attribute value
no attribute value
no attribute value
unknown Parameter value kind prim::Constant
no attribute value
no attribute value
no attribute value
no attribute value
unknown Parameter value kind prim::Constant
no attribute value
no attribute value
no attribute value
no attribute value
############# pass_level2
############# pass_level3
############# pass_level4
############# pass_level5
############# pass_ncnn

model | 模型 | モデル

1. original model

how to reproduce | 复现步骤 | 再現方法

1.下载GFPGANCleanv1-NoCE-C2,网址 https://github.com/TencentARC/GFPGAN/releases/download/v0.2.0/GFPGANCleanv1-NoCE-C2.pth
2.pythor下保存track文件，示例代码如下

import cv2
import os
import torch
from basicsr.utils import img2tensor, tensor2img
from basicsr.utils.download_util import load_file_from_url
from facexlib.utils.face_restoration_helper import FaceRestoreHelper
from torchvision.transforms.functional import normalize

from gfpgan.archs.gfpganv1_arch import GFPGANv1
from gfpgan.archs.gfpganv1_clean_arch import GFPGANv1Clean

ROOT_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))


def main():
    device = torch.device('cuda' if not torch.cuda.is_available() else 'cpu')
    gfpgan = GFPGANv1Clean(
        out_size=512,
        num_style_feat=512,
        channel_multiplier=2,
        decoder_load_path=None,
        fix_decoder=False,
        num_mlp=8,
        input_is_latent=True,
        different_w=True,
        narrow=1,
        sft_half=True)

    cropped_face = cv2.imread("E://workspace//GFPGAN-master//results//cropped_faces//00_01.png")
    cropped_face_t = img2tensor(cropped_face / 255., bgr2rgb=True, float32=True)
    normalize(cropped_face_t, (0.5, 0.5, 0.5), (0.5, 0.5, 0.5), inplace=True)
    cropped_face_t = cropped_face_t.unsqueeze(0).to(device)

    model_path = "..//experiments//pretrained_models//GFPGANCleanv1-NoCE-C2.pth"
    loadnet = torch.load(model_path)
    gfpgan.load_state_dict(loadnet['params_ema'], strict=True)
    gfpgan.eval()
    gfpgan = gfpgan.to(device)

    output = gfpgan(cropped_face_t, return_rgb=False)[0]
    restored_face = tensor2img(output.squeeze(0), rgb2bgr=True, min_max=(-1, 1))

    cv2.imshow("cropped_face_t", cropped_face)
    cv2.imshow("restored_face", restored_face)

    cv2.waitKey(2000)

    print("+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++")

    jit_track = torch.jit.trace(gfpgan, cropped_face_t)
    torch.jit.save(jit_track, "GFPGANCleanv1-NoCE-C2.trace")

    pass


if __name__ == "__main__":
    main()

3.pnnx.exe GFPGANCleanv1-NoCE-C2.trace inputshape=[1,3,512,512]

其他：
环境Window10 VS2017 x64 libtorch=1.10。

补充信息：
使用libtorch c++加载GFPGANCleanv1-NoCE-C2.trace可正常运行。
其他可能有用的信息如此连接描述 #3262 (comment)

@nihui

[nihui]

fixed in e0124db

感谢反馈

[zhu-zhaofei]

感谢大佬FIXED.