项目地址:https://github.com/lutzroeder/Netro
参考:使用netron工具可视化pytorch模型_jieleiping的博客-CSDN博客_netron pytorch
1.安装netron
pip install netron
2.直接可视化onnx/pth/pt(注意部分版本pt/pth需要转化为onnx,因为出现没线)
import netron
netron.start('net_pytorch.pth')
netron.start('net.onnx')
3.模型导出为onnx格式
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.onnx
import netron
class ForwardNet(nn.Module):
def __init__(self):
super(ForwardNet, self).__init__()
self.block1 = nn.Sequential(
nn.Conv2d(64, 64, 3, padding=1, bias=False),
nn.BatchNorm2d(64),
nn.ReLU(inplace=True),
nn.Conv2d(64, 32, 1, bias=False),
nn.BatchNorm2d(32),
nn.ReLU(inplace=True),
nn.Conv2d(32, 64, 3, padding=1, bias=False),
nn.BatchNorm2d(64)
)
self.conv1 = nn.Conv2d(3, 64, 3, padding=1, bias=False)
self.output = nn.Sequential(
nn.Conv2d(64, 1, 3, padding=1, bias=True),
nn.Sigmoid()
)
def forward(self, x):
x = self.conv1(x)
identity = x
x = F.relu(self.block1(x) + identity)
x = self.output(x)
return x
input = torch.rand(1, 3, 416, 416)
model = ForwardNet()
output = model(input)
onnx_path = "netForwatch.onnx"
torch.onnx.export(model, input, onnx_path)
netron.start(onnx_path)
执行上面代码后,会调用本地浏览器打开,形式和tensorboard差不多。
4.pth/pt转onnx
对pytorch模型格式(.pt/.pth)支持不友好,因此需要存为onnx
import torch
from model import Model
pytorch_net_path = 'net.pth'
onnx_net_path = 'net.onnx'
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu' )
# 权重导入模型
net = Model().to(device)
net.load_state_dict(torch.load(pytorch_net_path, map_location=device))
net.eval()
input = torch.randn(1, 1, 30, 30).to(device) # BCHW 其中Batch必须为1,因为测试时一般为1,尺寸HW必须和训练时的尺寸一致
torch.onnx.export(net, input, onnx_net_path, verbose=False)
参考:PyTorch模型训练特征图可视化(TensorboardX) - 知乎