module 'torch' has no attribute '_TensorBase' about simple-faster-rcnn-pytorch

[dami23]

I have met the same error. The solution is change the t._TensorBase to t._C._TensorBase

[rxqy]

Also, I'm guessing another problem you will gonna encounter is something like 'Can't call numpy() on Variable'
I changed line 11 from array_tool.py
if isinstance(data, t._C._TensorBase):
to
if isinstance(data, t._C._TensorBase) and (data.requires_grad==False):
I don't know if it's an error with pytorch or not, but it works for now

[rxqy]

I am also encountering a similar problem here.
I followed @dami23 , then changed the line [ loc_loss /= (gt_label >= 0).sum() ] to [ loc_loss /= (gt_label >= 0).sum().float() ].

[woxihuanchirou]

I cant find anything about torch._TensorBase,but torch.cuda.LongTensor,so I change utils/array_tool.py in line 43:
if isinstance(data, t._TensorBase):
to
if isinstance(data, t.cuda.LongTensor):
it occured that:RecursionError: maximum recursion depth exceeded
How to solve it?

[woxihuanchirou]

@dami23 thank you for replying.
I have changed it already,but a new error occured like this:

Traceback (most recent call last):
  File "train.py", line 131, in <module>
    fire.Fire()
  File "/home/htu/.local/lib/python3.5/site-packages/fire/core.py", line 127, in Fire
    component_trace = _Fire(component, args, context, name)
  File "/home/htu/.local/lib/python3.5/site-packages/fire/core.py", line 366, in _Fire
    component, remaining_args)
  File "/home/htu/.local/lib/python3.5/site-packages/fire/core.py", line 542, in _CallCallable
    result = fn(*varargs, **kwargs)
  File "train.py", line 81, in train
    trainer.train_step(img, bbox, label, scale)
  File "/home/htu/yc/simple-faster-rcnn-pytorch-master/trainer.py", line 168, in train_step
    losses = self.forward(imgs, bboxes, labels, scale)
  File "/home/htu/yc/simple-faster-rcnn-pytorch-master/trainer.py", line 135, in forward
    self.rpn_sigma)
  File "/home/htu/yc/simple-faster-rcnn-pytorch-master/trainer.py", line 253, in _fast_rcnn_loc_loss
    loc_loss /= (gt_label >= 0).sum()  # ignore gt_label==-1 for rpn_loss
RuntimeError: Expected object of type torch.cuda.FloatTensor but found type torch.cuda.LongTensor for argument #3 'other'

[woxihuanchirou]

@rxqy thank you so much!
I encountered a new problem like this:

Traceback (most recent call last):
  File "train.py", line 131, in <module>
    fire.Fire()
  File "/home/htu/.local/lib/python3.5/site-packages/fire/core.py", line 127, in Fire
    component_trace = _Fire(component, args, context, name)
  File "/home/htu/.local/lib/python3.5/site-packages/fire/core.py", line 366, in _Fire
    component, remaining_args)
  File "/home/htu/.local/lib/python3.5/site-packages/fire/core.py", line 542, in _CallCallable
    result = fn(*varargs, **kwargs)
  File "train.py", line 65, in train
    faster_rcnn = FasterRCNNVGG16()
  File "/home/htu/yc/simple-faster-rcnn-pytorch-master/model/faster_rcnn_vgg16.py", line 62, in __init__
    extractor, classifier = decom_vgg16()
  File "/home/htu/yc/simple-faster-rcnn-pytorch-master/model/faster_rcnn_vgg16.py", line 16, in decom_vgg16
    model.load_state_dict(t.load(opt.caffe_pretrain_path))
  File "/usr/local/lib/python3.5/dist-packages/torch/serialization.py", line 267, in load
    return _load(f, map_location, pickle_module)
  File "/usr/local/lib/python3.5/dist-packages/torch/serialization.py", line 420, in _load
    result = unpickler.load()
AttributeError: Can't get attribute '_rebuild_tensor_v2' on <module 'torch._utils' from '/usr/local/lib/python3.5/dist-packages/torch/_utils.py'>

That's because the version of Pytorch is too old.https://github.com/pytorch/pytorch/issues/5729
I have encountered so many problems about packages version. So could you tell me what's the version of your packages like torch,torchvision,tronado and cupy.
Here is mine:
cupy-cuda80 4.0.0
torch 0.4.0
torchfile 0.1.0
torchnet 0.0.2
torchvision 0.2.1
tornado 4.5.3

[rxqy]

Name Version Build Channel
cupy-cuda80 4.0.0
pytorch 0.4.0 py36_cuda8.0.61_cudnn7.1.2_1 pytorch
torchfile 0.1.0
torchnet 0.0.2
torchvision 0.2.1 py36_1 pytorch
tornado 4.5.3 py36_0

I'm using most of them from the latest version in anaconda3.

Did you run commands like
pip install -r requirements.txt ?
I think that'll solve most dependency problems

[woxihuanchirou]

@rxqy I have already run pip install -r requirements.txt without errors.Then I run python3 build.py build_ext --inplace,a warning occured like this:

running build_ext
cythoning _nms_gpu_post.pyx to _nms_gpu_post.c
building '_nms_gpu_post' extension
creating build
creating build/temp.linux-x86_64-3.5
x86_64-linux-gnu-gcc -pthread -DNDEBUG -g -fwrapv -O2 -Wall -Wstrict-prototypes -g -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2 -fPIC -I/usr/include/python3.5m -c _nms_gpu_post.c -o build/temp.linux-x86_64-3.5/_nms_gpu_post.o
In file included from /usr/include/numpy/ndarraytypes.h:1777:0,
                 from /usr/include/numpy/ndarrayobject.h:18,
                 from /usr/include/numpy/arrayobject.h:4,
                 from _nms_gpu_post.c:566:
/usr/include/numpy/npy_1_7_deprecated_api.h:15:2: warning: #warning "Using deprecated NumPy API, disable it by " "#defining NPY_NO_DEPRECATED_API NPY_1_7_API_VERSION" [-Wcpp]
 #warning "Using deprecated NumPy API, disable it by " \
  ^
x86_64-linux-gnu-gcc -pthread -shared -Wl,-O1 -Wl,-Bsymbolic-functions -Wl,-Bsymbolic-functions -Wl,-z,relro -Wl,-Bsymbolic-functions -Wl,-z,relro -g -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2 build/temp.linux-x86_64-3.5/_nms_gpu_post.o -o /home/htu/yc/simple-faster-rcnn-pytorch-master/model/utils/nms/_nms_gpu_post.cpython-35m-x86_64-linux-gnu.so

I think it won't trouble me,yes?
Then I run python3 train.py train --env='fasterrcnn_caffe' --plot_every=100 --caffe_pretrain,the AttributeError occured again~
It really troubles me a lot! The version of my Python is 3.5.2.Is it because of this problem?

[rxqy]

Sorry but I didn't build this gpu version of nms, I just skipped that step.
About the attribute error, you might want to refer to this link from pytorch forum.
https://discuss.pytorch.org/t/question-about-rebuild-tensor-v2/14560/4

I think all these errors are coming from the different versions of pytorch....
Maybe switch to a 0.3 may help..

[zhuotest]

The reference url is https://discuss.pytorch.org/t/question-about-rebuild-tensor-v2/14560/2

Putting the following line just after import torch. Works for me. torch 0.4, python3.

# Monkey-patch because I trained with a newer version.
# This can be removed once PyTorch 0.4.x is out.
# See https://discuss.pytorch.org/t/question-about-rebuild-tensor-v2/14560
import torch._utils
try:
    torch._utils._rebuild_tensor_v2
except AttributeError:
    def _rebuild_tensor_v2(storage, storage_offset, size, stride, requires_grad, backward_hooks):
        tensor = torch._utils._rebuild_tensor(storage, storage_offset, size, stride)
        tensor.requires_grad = requires_grad
        tensor._backward_hooks = backward_hooks
        return tensor
    torch._utils._rebuild_tensor_v2 = _rebuild_tensor_v2

[perry0418]

In my case, under the torch 0.4.1, I occur another similar error:
File "/home/ubuntu/anaconda3/envs/pytorch/lib/python3.6/site-packages/torchnet/meter/averagevaluemeter.py", line 29, in add
self.std = np.sqrt(self.m_s / (self.n - 1.0))
File "/home/ubuntu/anaconda3/envs/pytorch/lib/python3.6/site-packages/torch/tensor.py", line 397, in array
return self.cpu().numpy()
RuntimeError: Can't call numpy() on Variable that requires grad. Use var.detach().numpy() instead.

As disscussed in this link: https://github.com/pytorch/pytorch/issues/7610
numpy tries to call the array in tensor.py and the raise the error

so I change the array in site-packages/torch/tensor.py in line 397 as follows:
def __array__(self, dtype=None): if dtype is None: if (self.requires_grad==False): return self.cpu().numpy() else: return self.data.cpu().numpy() else: return self.cpu().numpy().astype(dtype, copy=False)

It works for me, and it may be helpful for others.