xiezheng-cs / pytorchtocaffe Goto Github PK

最好是使用作者开发的模块 https://github.com/hujie-frank/SENet
具体思路可以分享一下吗？

你好，我现在能将FeatherNet的pytorh模型生成caffemodel和prototxt，但是prototxt好像不对，前向的时候会报错。view2维度是16x1x1，但是add1的维度是16x56x56，这两个mul1 做运算的时候会报错

你好，我正好也要使用FeatherNet 的caffe模型， 我从FeatherNet作者那里下载了模型，他那里的图片输入是224x224，用你的工具尝试转化没有成功。我看到你的FeatherNet模型的输入是112x112，应该成功了吧，可以开源出来吗

import torch
from torchvision.models import resnet
import pytorch_to_caffe
from facenet_pytorch import InceptionResnetV1

if name=='main':
name='VGGFace2'
# For a model pretrained on VGGFace2
model = InceptionResnetV1(pretrained='vggface2').eval()

input=torch.ones([1, 3, 512, 512])

pytorch_to_caffe.trans_net(model, input, name)
pytorch_to_caffe.save_prototxt('{}.prototxt'.format(name))
pytorch_to_caffe.save_caffemodel('{}.caffemodel'.format(name))

Add blob conv_blob13 : torch.Size([1, 256, 61, 61])
140602027406272:cat_blob1 getting
140602027407552:conv_blob13 getting
mul1 was added to layers
140602027406752:mul_blob1 was added to blobs
Add blob mul_blob1 : torch.Size([1, 256, 61, 61])
140602027407552:conv_blob13 getting
Traceback (most recent call last):
File "VGGFace2_pytorch_2_caffe.py", line 14, in
pytorch_to_caffe.trans_net(model, input, name)
File "/home/songdc/Documents/PytorchToCaffe/pytorch_to_caffe.py", line 675, in trans_net
out = net.forward(input_var)
File "/home/songdc/miniconda3/lib/python3.7/site-packages/facenet_pytorch/models/inception_resnet_v1.py", line 288, in forward
x = self.repeat_1(x)
File "/home/songdc/miniconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 489, in call
result = self.forward(*input, **kwargs)
File "/home/songdc/miniconda3/lib/python3.7/site-packages/torch/nn/modules/container.py", line 92, in forward
input = module(input)
File "/home/songdc/miniconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 489, in call
result = self.forward(*input, **kwargs)
File "/home/songdc/miniconda3/lib/python3.7/site-packages/facenet_pytorch/models/inception_resnet_v1.py", line 65, in forward
out = out * self.scale + x
File "/home/songdc/Documents/PytorchToCaffe/pytorch_to_caffe.py", line 552, in _mul
bottom=[log.blobs(input), log.blobs(args[0])], top=top_blobs)
File "/home/songdc/Documents/PytorchToCaffe/pytorch_to_caffe.py", line 101, in blobs
print("{}:{} getting".format(var, self._blobs[var]))
File "/home/songdc/Documents/PytorchToCaffe/pytorch_to_caffe.py", line 39, in getitem
return self.data[key]
KeyError: 140602226322992

torch：1.0
torchvision：0.2
（haven add code torchvison.ops.boxes to facenet_pytorch source code which occur in torchvision>0.3）

Why does the error occur？ Thank you！

HI, I successed in dumping out the prototxt and caffemodel ,I am having problems when loading prototxt and caffemodel in pycaffe interface ,

or have you met the same problems ,thanks in advanced