heromanba / 3d-r2n2-pytorch Goto Github PK

这个是自己创建的吗？新手求教

why you didnt consider the IOU mentioned in the paper

Dear author, could you provide us with information about the training that was done to obtain the pretrained model(ResidualGRUNet), such as the configuration used, the hardware, etc.
Thanks in advance.

I had the following problems during training, I guess there are some problem in the loop in the forward function in base_gru_net. I'm looking forward to your help!

Traceback (most recent call last):
File "main.py", line 121, in
main()
File "main.py", line 111, in main
train_net()
File "/home/may/下载/3D-R2N2-PyTorch-master/lib/train_net.py", line 22, in func_wrapper
return func(*args, **kwargs)
File "/home/may/下载/3D-R2N2-PyTorch-master/lib/train_net.py", line 75, in train_net
solver.train(train_queue, val_queue)
File "/home/may/下载/3D-R2N2-PyTorch-master/lib/solver.py", line 113, in train
loss = self.train_loss(batch_img, batch_voxel)
File "/home/may/下载/3D-R2N2-PyTorch-master/lib/solver.py", line 67, in train_loss
loss = self.net(x, y, test=False)
File "/home/may/anaconda2/envs/py3torch/lib/python3.6/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "/home/may/下载/3D-R2N2-PyTorch-master/models/base_gru_net.py", line 76, in forward
gru_out, update_gate = self.encoder(x[time], h, u, time)
File "/home/may/anaconda2/envs/py3torch/lib/python3.6/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
TypeError: forward() takes 4 positional arguments but 5 were given
[INFO/MainProcess] process shutting down
[INFO/MainProcess] calling join() for process ReconstructionDataProcess-6

Thank you for your work.When I test on your code ,I do not understand why the second parameter is 2.

When the input is torch.randn(5, 1, 3, 127, 127), I get the result [1,2,32,32,32] ,I do not understand why the second parameter of result is 2.Waiting for your reply!

有没有其他地址，下载速度过分漫长。
'curl', 'ftp://cs.stanford.edu/cs/cvgl/ResidualGRUNet.npy',

谢谢

Hi, thank you for your wonderful work. I can already train. I would like to ask some questions about the pre-trained model. First, is the model obtained by training all categories of Shapenet? I tested the reconstruction effect of a single plane and it was not ideal. Second, whether to output the model file during training. Third, whether to provide an evaluation code.

Looking forward to your reply

Drive link is not working, neither I have a biadu account. How can I download the weights?

I tried passing a 2D image of the plane from PASCAL VOC dataset to replicate the results shown in Fig. 4a. The image attached below was passed. However, while running demo.py, the result is not as expected.

The output generated looks like this which is nowhere near to the structure seen in the 2D image.

Is there anything that needs to be taken care of? Thanks in advance!

Update: in the paper, the network has been fine-tuned on PASCAL 3D+ dataset. Are those weights available?

The following link does not work anymore.

Google drive link(https://drive.google.com/open?id=1LtNhuUQdAeAyIUiuCavofBpjw26Ag6DP)

When I was training, the following problems occurred in the middle of my training.
looking forward to your reply.

2020-09-08 09:50:59.346583 Iter: 2880 Loss: 0.171822
Wait until the dataprocesses to end
Traceback (most recent call last):
File "main.py", line 121, in
main()
File "main.py", line 111, in main
train_net()
File "/home/zyz/Project/3D-R2N2-PyTorch/lib/train_net.py", line 24, in func_wrapper
return func(*args, **kwargs)
File "/home/zyz/Project/3D-R2N2-PyTorch/lib/train_net.py", line 84, in train_net
solver.train(train_loader, val_loader)
File "/home/zyz/Project/3D-R2N2-PyTorch/lib/solver.py", line 117, in train
loss = self.train_loss(batch_img, batch_voxel)
File "/home/zyz/Project/3D-R2N2-PyTorch/lib/solver.py", line 65, in train_loss
loss = self.net(x, y, test=False)
File "/home/zyz/anaconda3/envs/3D-R2N2/lib/python3.6/site-packages/torch/nn/modules/module.py", line 532, in call
result = self.forward(*input, **kwargs)
File "/home/zyz/Project/3D-R2N2-PyTorch/models/base_gru_net.py", line 76, in forward
gru_out, update_gate = self.encoder(x[time], h, u, time)
File "/home/zyz/anaconda3/envs/3D-R2N2/lib/python3.6/site-packages/torch/nn/modules/module.py", line 532, in call
result = self.forward(*input, **kwargs)
File "/home/zyz/Project/3D-R2N2-PyTorch/models/res_gru_net.py", line 155, in forward
t_x_s_update = self.t_x_s_update(rect7, h)
File "/home/zyz/anaconda3/envs/3D-R2N2/lib/python3.6/site-packages/torch/nn/modules/module.py", line 532, in call
result = self.forward(*input, **kwargs)
File "/home/zyz/Project/3D-R2N2-PyTorch/lib/layers.py", line 45, in forward
out = self.fc_layer(fc7).view(*target_shape) + self.conv3d(h) + self.bias
RuntimeError: The size of tensor a (5) must match the size of tensor b (12) at non-singleton dimension 0

Traceback (most recent call last):
File "demo.py", line 90, in
main()
File "demo.py", line 67, in main
net.load(DEFAULT_WEIGHTS) # load downloaded weights
File "/home/zhangq/anaconda3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 518, in getattr
type(self).name, name))
AttributeError: 'ResidualGRUNet' object has no attribute 'load'

在自己训练运行代码是出现这种情况一般是什么原因，我没有改参数，我还看不懂
还需要问一下，怎样学习看这种代码，从哪个文件开始

hello，作者你好，如果我想对真实物体进行重建尝试，比如在白色幕布面前拍摄的椅子等图片。
请问您知道应该采用什么样的方式对真实照片进行预处理吗？

作者能否加下你的联系方式，想了解这个项目更多的细节，有偿