Vision: RuntimeError: builtin cannot be used as a value

[marting87]

On commit c7c2085ec686ccc55e1df85736b240b2405d1179 I'm receiving the following error whenever i call import torchvision.models as models:

RuntimeError: 
builtin cannot be used as a value:
at venv/lib/python3.6/site-packages/torchvision-0.5.0a0+c7c2085-py3.6-linux-x86_64.egg/torchvision/models/detection/_utils.py:14:56
def zeros_like(tensor, dtype):
    # type: (Tensor, int) -> Tensor
    return torch.zeros_like(tensor, dtype=dtype, layout=tensor.layout,
                                                        ~~~~~~~~~~~~~ <--- HERE
                            device=tensor.device, pin_memory=tensor.is_pinned())
'zeros_like' is being compiled since it was called from '__torch__.torchvision.models.detection._utils.BalancedPositiveNegativeSampler.__call__'
at venv/lib/python3.6/site-packages/torchvision-0.5.0a0+c7c2085-py3.6-linux-x86_64.egg/torchvision/models/detection/_utils.py:72:12

            # randomly select positive and negative examples
            perm1 = torch.randperm(positive.numel(), device=positive.device)[:num_pos]
            perm2 = torch.randperm(negative.numel(), device=negative.device)[:num_neg]

            pos_idx_per_image = positive[perm1]
            neg_idx_per_image = negative[perm2]

            # create binary mask from indices
            pos_idx_per_image_mask = zeros_like(
            ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~...  <--- HERE
                matched_idxs_per_image, dtype=torch.uint8
            )
            neg_idx_per_image_mask = zeros_like(
                matched_idxs_per_image, dtype=torch.uint8
            )

            pos_idx_per_image_mask[pos_idx_per_image] = torch.tensor(1, dtype=torch.uint8)
            neg_idx_per_image_mask[neg_idx_per_image] = torch.tensor(1, dtype=torch.uint8)

[fmassa]

Thanks for the bug report!

What's your PyTorch version?

We might need to make a backwards-compatible fix for version PyTorch 1.4 if this isn't supported there.

@eellison do know you if tensor.layout is supported in the 1.4 branch?

[marting87]

My PyTorch version = 1.3.1 (latest from pip)

[fmassa]

Yeah, that's expected -- you'll need a more recent version of PyTorch (at least from September, containing https://github.com/pytorch/pytorch/commit/6a4ca9abec1c18184635881c08628737c8ed2497)

@eellison told me that this patch is going to be present in the 1.4 release, so the solution here is just to update PyTorch to a nightly, or wait for 1.4 to be released in early January.

Thanks for the heads up!

[marting87]

@fmassa Thanks! I'll switch to nigthly for now

[martinlightmatter]

I just ran into this when using torch==1.4.0 and torchvision==0.5.0. Rolling back to torch==1.3.1 and `torchvision==0.4.2" seems to fix the issue for me.

python environment details:
Python 3.7.5
torchvision 0.5.0
torch 1.4.0

host details:
Virtualization: microsoft
Operating System: Ubuntu 18.04.3 LTS
Kernel: Linux 5.0.0-1031-azure
Architecture: x86-64

traceback:

  File "./py37/lib/python3.7/site-packages/torchvision/__init__.py", line 3, in <module>
    from torchvision import models
  File "./py37/lib/python3.7/site-packages/torchvision/models/__init__.py", line 12, in <module>
    from . import detection
  File "./py37/lib/python3.7/site-packages/torchvision/models/detection/__init__.py", line 1, in <module>
    from .faster_rcnn import *
  File "./py37/lib/python3.7/site-packages/torchvision/models/detection/faster_rcnn.py", line 13, in <module>
    from .rpn import AnchorGenerator, RPNHead, RegionProposalNetwork
  File "./py37/lib/python3.7/site-packages/torchvision/models/detection/rpn.py", line 11, in <module>
    from . import _utils as det_utils
  File "./py37/lib/python3.7/site-packages/torchvision/models/detection/_utils.py", line 19, in <module>
    class BalancedPositiveNegativeSampler(object):
  File "./py37/lib/python3.7/site-packages/torch/jit/__init__.py", line 1219, in script
    _compile_and_register_class(obj, _rcb, qualified_name)
  File "./py37/lib/python3.7/site-packages/torch/jit/__init__.py", line 1076, in _compile_and_register_class
    _jit_script_class_compile(qualified_name, ast, rcb)
  File "./py37/lib/python3.7/site-packages/torch/jit/_recursive.py", line 222, in try_compile_fn
    return torch.jit.script(fn, _rcb=rcb)
  File "./py37/lib/python3.7/site-packages/torch/jit/__init__.py", line 1226, in script
    fn = torch._C._jit_script_compile(qualified_name, ast, _rcb, get_default_args(obj))
RuntimeError:
builtin cannot be used as a value:
at ./py37/lib/python3.7/site-packages/torchvision/models/detection/_utils.py:14:56
def zeros_like(tensor, dtype):
    # type: (Tensor, int) -> Tensor
    return torch.zeros_like(tensor, dtype=dtype, layout=tensor.layout,
                                                        ~~~~~~~~~~~~~ <--- HERE
                            device=tensor.device, pin_memory=tensor.is_pinned())
'zeros_like' is being compiled since it was called from '__torch__.torchvision.models.detection._utils.BalancedPositiveNegativeSampler.__call__'
at ./py37/lib/python3.7/site-packages/torchvision/models/detection/_utils.py:72:12

            # randomly select positive and negative examples
            perm1 = torch.randperm(positive.numel(), device=positive.device)[:num_pos]
            perm2 = torch.randperm(negative.numel(), device=negative.device)[:num_neg]

            pos_idx_per_image = positive[perm1]
            neg_idx_per_image = negative[perm2]

            # create binary mask from indices
            pos_idx_per_image_mask = zeros_like(
            ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~...  <--- HERE
                matched_idxs_per_image, dtype=torch.uint8
            )
            neg_idx_per_image_mask = zeros_like(
                matched_idxs_per_image, dtype=torch.uint8
            )

            pos_idx_per_image_mask[pos_idx_per_image] = torch.tensor(1, dtype=torch.uint8)
            neg_idx_per_image_mask[neg_idx_per_image] = torch.tensor(1, dtype=torch.uint8)

[fmassa]

@martinlightmatter this is weird, my first guess would be that you have a conflict of libraries in your system and it's picking the wrong combination of PyTorch / torchvision.

Indeed, while looking at your stacktrace, it points to

File "./py37/lib/python3.7/site-packages/torch/jit/_recursive.py", line 222, in try_compile_fn
    return torch.jit.script(fn, _rcb=rcb)

which, in PyTorch version 1.4.0, doesn't correspond to the same function, see https://github.com/pytorch/pytorch/blob/7f73f1d591afba823daa4a99a939217fb54d7688/torch/jit/_recursive.py#L222

Can you double-check your environment?

[AlexXXEL]

It seems that torchvision-0.4 will match pytorch-1.4

[fmassa]

@AlexXXEL torchvision 0.5 matches PyTorch 1.4, and they should come together otherwise it won't work properly

[AlexXXEL]

ok，thank
---原始邮件---
发件人:"Francisco Massa"[e-mail]
发送时间:2020年3月11日 星期三 下午10:6
收件人:"pytorch/vision"[e-mail]
抄送:"AlexXXEL"[e-mail];"Mention"[e-mail]
主题:Re: [pytorch/vision] RuntimeError: builtin cannot be used as a value(#1675)

@AlexXXELtorchvision 0.5 matches PyTorch 1.4, and they should come together otherwise it won't work properly

—
You are receiving this because you were mentioned.
Reply to this email directly,view it on GitHub, orunsubscribe.