How to convert the pytorch model to the onnx model?

[DidaDidaDidaD]

How to convert the pytorch model to the onnx model? I tried the conversion process, but I reported an error. I don't know what the problem is. I'm Xiaobai. Thank you for your advice.My script as follows: import torch import importlib

device = torch.device("cpu") model = "e2fgvi_hq"

ckpt = 'release_model/E2FGVI-HQ-CVPR22.pth'

net = importlib.import_module('model.' + model) model = net.InpaintGenerator().to(device) data = torch.load(ckpt, map_location=device) model.load_state_dict(data) print(f'Loading model from: {ckpt}') model.eval() x = torch.randn(1,1, 3, 240, 864, requires_grad=True) torch.onnx.export(model, # model being run (x,2), # model input (or a tuple for multiple inputs) "E2FGVI-HQ-CVPR22.onnx", # where to save the model (can be a file or file-like object) export_params=True, # store the trained parameter weights inside the model file opset_version=16, # the ONNX version to export the model to do_constant_folding=True, # whether to execute constant folding for optimization input_names = ['input'], # the model's input names output_names = ['output'], # the model's output names dynamic_axes={'input' : {1 : 'batch_size'}}) the error as follows: torch.onnx.symbolic_registry.UnsupportedOperatorError: Exporting the operator ::col2im to ONNX opset version 16 is not supported. Please feel free to request support or submit a pull request on PyTorch GitHub.

[magicse]

@DidaDidaDidaD pth to pt

import sys
sys.path.append('./')

import importlib

import torch
import torchvision
from model.e2fgvi_hq import InpaintGenerator
model_path = "E2FGVI-HQ-CVPR22.pth"

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = "e2fgvi_hq"
net = importlib.import_module('model.' + model)

model = net.InpaintGenerator().to(device)

data = torch.load(model_path, map_location=device)
model.load_state_dict(data)  

model.eval()

example1 = torch.rand(1, 3, 512, 512)
num_local_frames = 1

traced_script_module = torch.jit.trace(model, (example1, num_local_frames))
traced_script_module.save("E2FGVI-HQ-CVPR22.pt")

or like this

import sys
sys.path.append('./')

import importlib

import torch
import torchvision
from model.e2fgvi_hq import InpaintGenerator
from torch.jit import trace, ScriptModule, ignore

model_path = "E2FGVI-HQ-CVPR22.pth"


device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = "e2fgvi_hq"
net = importlib.import_module('model.' + model)

model = net.InpaintGenerator().to(device)

data = torch.load(model_path, map_location=device)
model.load_state_dict(data)  

# Switch the model to eval model
model.eval()

# An example input you would normally provide to your model's forward() method.
h = 480
w = 640
ch = 3
b = 1
t = 1 
imgs = torch.rand(t, b, ch, h, w).to(device)
print ("imgs shape", imgs.shape)
masked_imgs = imgs[:1, :1, :, :, :]
print ("masked imgs shape", masked_imgs.shape)

mod_size_h = 60
mod_size_w = 108
h_pad = (mod_size_h - h % mod_size_h) % mod_size_h
w_pad = (mod_size_w - w % mod_size_w) % mod_size_w
masked_imgs = torch.cat([masked_imgs, torch.flip(masked_imgs, [3])], 3)[:, :, :, :h + h_pad, :]
masked_imgs = torch.cat([masked_imgs, torch.flip(masked_imgs, [4])], 4)[:, :, :, :, :w + w_pad]
print ("masked imgs shape", masked_imgs.shape)

num_local_frames = torch.tensor(1)  # Set the desired value for num_local_frames
masked_frames = masked_imgs

traced_script_module = torch.jit.trace(model, (masked_frames, num_local_frames))

# Save the TorchScript model
traced_script_module.save("E2FGVI-HQ-CVPR22.pt")

after that use pnnx.exe

pnnx.exe E2FGVI-HQ-CVPR22.pt inputshape=[1,3,480,648],[1]  device=cpu

or

pnnx.exe E2FGVI-HQ-CVPR22.pt 
pnnx.exe E2FGVI-HQ-CVPR22.pt moduleop=model.modules.flow_comp.SPyNet,model.modules.flow_comp.SPyNetBasicModule

[eisneim]

@DidaDidaDidaD have you tried @magicse 's method? does it work?