Could not run '_caffe2::DistributeFpnProposals' with arguments from the 'CUDATensorId' backend. '

1. what changes you made (git diff) or what code you wrote i write a export_model.py

from detectron2.export.api import add_export_config, export_caffe2_model, Caffe2Model
from detectron2.config import get_cfg
from detectron2.modeling import build_model
from detectron2.engine.defaults import DefaultPredictor
from detectron2.data.detection_utils import read_image
import detectron2.data.transforms as T
import torch
import cv2
import os

def prepareConfig(testThreshold=0.8):
    # cfg
    config_file = 'configs/PascalVOC-Detection/faster_rcnn_R_50_FPN.yaml'
    cfg = get_cfg()
    cfg.merge_from_file(config_file)
    # Set score_threshold for builtin models
    cfg.MODEL.RETINANET.SCORE_THRESH_TEST = testThreshold
    cfg.MODEL.ROI_HEADS.SCORE_THRESH_TEST = testThreshold
    cfg.MODEL.PANOPTIC_FPN.COMBINE.INSTANCES_CONFIDENCE_THRESH = testThreshold
    cfg.MODEL.WEIGHTS = os.path.join('output/R50_FPN', 'model_final.pth')
    cfg.MODEL.ROI_HEADS.NUM_CLASSES = 20
    return cfg

cfg = prepareConfig()
cfg = add_export_config(cfg)
model = build_model(cfg)

img = read_image('1008.bmp', format='BGR')
height, width = img.shape[:2]
transform_gen = T.ResizeShortestEdge(
            [cfg.INPUT.MIN_SIZE_TEST, cfg.INPUT.MIN_SIZE_TEST], cfg.INPUT.MAX_SIZE_TEST
        )
image = transform_gen.get_transform(img).apply_image(img)
image = torch.as_tensor(image.astype("float32").transpose(2, 0, 1))

inputs = {'image':image, 'height':height, 'width':width}

c2_model = export_caffe2_model(cfg, model, [inputs])
c2_model.save_protobuf('./c2res')

1. what exact command you run: python demo/export_model.py
2. what you observed (including the full logs):

net_drawer will not run correctly. Please install the correct dependencies.
D:\Anaconda3\lib\site-packages\pycocotools\coco.py:49: UserWarning:
This call to matplotlib.use() has no effect because the backend has already
been chosen; matplotlib.use() must be called *before* pylab, matplotlib.pyplot,
or matplotlib.backends is imported for the first time.

The backend was *originally* set to 'Qt5Agg' by the following code:
  File "demo/export_model.py", line 1, in <module>
    from detectron2.export.api import add_export_config, export_caffe2_model, Caffe2Model
  File "d:\algorithmlibrary\detectron2-master\detectron2\export\__init__.py", line 3, in <module>
    from .api import *
  File "d:\algorithmlibrary\detectron2-master\detectron2\export\api.py", line 10, in <module>
    from .caffe2_inference import ProtobufDetectionModel
  File "d:\algorithmlibrary\detectron2-master\detectron2\export\caffe2_inference.py", line 11, in <module>
    from detectron2.export.caffe2_modeling import convert_batched_inputs_to_c2_format
  File "d:\algorithmlibrary\detectron2-master\detectron2\export\caffe2_modeling.py", line 9, in <module>
    from detectron2.modeling import meta_arch
  File "d:\algorithmlibrary\detectron2-master\detectron2\modeling\__init__.py", line 17, in <module>
    from .meta_arch import (
  File "d:\algorithmlibrary\detectron2-master\detectron2\modeling\meta_arch\__init__.py", line 9, in <module>
    from .rcnn import GeneralizedRCNN, ProposalNetwork
  File "d:\algorithmlibrary\detectron2-master\detectron2\modeling\meta_arch\rcnn.py", line 10, in <module>
    from detectron2.utils.visualizer import Visualizer
  File "d:\algorithmlibrary\detectron2-master\detectron2\utils\visualizer.py", line 13, in <module>
    from matplotlib.backends.backend_agg import FigureCanvasAgg
  File "D:\Anaconda3\lib\site-packages\matplotlib\backends\__init__.py", line 16, in <module>
    line for line in traceback.format_stack()


  import matplotlib; matplotlib.use('Agg')
Config 'configs/PascalVOC-Detection/faster_rcnn_R_50_FPN.yaml' has no VERSION. Assuming it to be compatible with latest v2.
d:\algorithmlibrary\detectron2-master\detectron2\export\c10.py:24: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  assert tensor.dim() == 2 and tensor.size(-1) in [4, 5], tensor.size()
Traceback (most recent call last):
  File "demo/export_model.py", line 39, in <module>
    c2_model = export_caffe2_model(cfg, predictor.model, [inputs])
  File "d:\algorithmlibrary\detectron2-master\detectron2\export\api.py", line 54, in export_caffe2_model
    predict_net, init_net = export_caffe2_detection_model(c2_compatible_model, c2_format_input)
  File "d:\algorithmlibrary\detectron2-master\detectron2\export\caffe2_export.py", line 80, in export_caffe2_detection_model
    predict_net, init_net = _export_via_onnx(model, (tensor_inputs,))
  File "d:\algorithmlibrary\detectron2-master\detectron2\export\caffe2_export.py", line 39, in _export_via_onnx
    f,
  File "D:\Anaconda3\lib\site-packages\torch\onnx\__init__.py", line 156, in export
    custom_opsets)
  File "D:\Anaconda3\lib\site-packages\torch\onnx\utils.py", line 67, in export
    custom_opsets=custom_opsets)
  File "D:\Anaconda3\lib\site-packages\torch\onnx\utils.py", line 466, in _export
    fixed_batch_size=fixed_batch_size)
  File "D:\Anaconda3\lib\site-packages\torch\onnx\utils.py", line 319, in _model_to_graph
    graph, torch_out = _trace_and_get_graph_from_model(model, args, training)
  File "D:\Anaconda3\lib\site-packages\torch\onnx\utils.py", line 276, in _trace_and_get_graph_from_model
    torch.jit._get_trace_graph(model, args, _force_outplace=False, _return_inputs_states=True)
  File "D:\Anaconda3\lib\site-packages\torch\jit\__init__.py", line 282, in _get_trace_graph
    outs = ONNXTracedModule(f, _force_outplace, return_inputs, _return_inputs_states)(*args, **kwargs)
  File "D:\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 539, in __call__
    result = self.forward(*input, **kwargs)
  File "D:\Anaconda3\lib\site-packages\torch\jit\__init__.py", line 365, in forward
    self._force_outplace,
  File "D:\Anaconda3\lib\site-packages\torch\jit\__init__.py", line 352, in wrapper
    outs.append(self.inner(*trace_inputs))
  File "D:\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 537, in __call__
    result = self._slow_forward(*input, **kwargs)
  File "D:\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 523, in _slow_forward
    result = self.forward(*input, **kwargs)
  File "D:\Anaconda3\lib\contextlib.py", line 52, in inner
    return func(*args, **kwds)
  File "d:\algorithmlibrary\detectron2-master\detectron2\export\caffe2_modeling.py", line 122, in forward
    results = self._wrapped_model.inference(inputs, do_postprocess=False)
  File "d:\algorithmlibrary\detectron2-master\detectron2\modeling\meta_arch\rcnn.py", line 171, in inference
    results, _ = self.roi_heads(images, features, proposals, None)
  File "D:\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 537, in __call__
    result = self._slow_forward(*input, **kwargs)
  File "D:\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 523, in _slow_forward
    result = self.forward(*input, **kwargs)
  File "d:\algorithmlibrary\detectron2-master\detectron2\modeling\roi_heads\roi_heads.py", line 572, in forward
    pred_instances = self._forward_box(features_list, proposals)
  File "d:\algorithmlibrary\detectron2-master\detectron2\modeling\roi_heads\roi_heads.py", line 619, in _forward_box
    box_features = self.box_pooler(features, [x.proposal_boxes for x in proposals])
  File "D:\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 537, in __call__
    result = self._slow_forward(*input, **kwargs)
  File "D:\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 523, in _slow_forward
    result = self.forward(*input, **kwargs)
  File "d:\algorithmlibrary\detectron2-master\detectron2\export\c10.py", line 298, in forward
    legacy_plus_one=False,
RuntimeError: Could not run '_caffe2::DistributeFpnProposals' with arguments from the 'CUDATensorId' backend. '_caffe2::DistributeFpnProposals' is only available for these backends: [CPUTensorId]. (dispatch_ at C:\w\1\s\windows\pytorch\aten\src\ATen/core/dispatch/Dispatcher.h:222)
(no backtrace available)

Environment:

sys.platform win32 Python 3.6.5 |Anaconda, Inc.| (default, Mar 29 2018, 13:32:41) [MSC v.1900 64 bit (AMD64)] Numpy 1.14.3 Detectron2 Compiler MSVC 191125548 Detectron2 CUDA Compiler 10.1 DETECTRON2_ENV_MODULE <not set> PyTorch 1.4.0.dev20200107 PyTorch Debug Build False torchvision 0.5.0.dev20200107 CUDA available True GPU 0 GeForce RTX 2080 Ti CUDA_HOME C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1 NVCC Not Available Pillow 6.0.0 cv2 4.1.1

---

PyTorch built with:

• MSVC 191125547
• Intel® Math Kernel Library Version 2019.0.4 Product Build 20190411 for Intel® 64 architecture applications
• OpenMP 200203
• CUDA Runtime 10.1
• NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_37,code=compute_37
• CuDNN 7.6.3
• Magma 2.5.1
• Build settings: BLAS=MKL, BUILD_TYPE=Release, CXX_FLAGS=/DWIN32 /D_WINDOWS /GR /w /EHa /MP /bigobj -openmp, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, USE_CUDA=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=OFF, USE_MPI=OFF, USE_NCCL=OFF, USE_NNPACK=OFF, USE_OPENMP=ON, USE_STATIC_DISPATCH=OFF,

[ppwwyyxx]

Our CPU inference is fast enough and much cheaper than GPUs.

[CodeSammich]

GPU export fails with the error RuntimeError: Could not run '_caffe2::GenerateProposals' with arguments from the 'CUDATensorId' backend. '_caffe2::GenerateProposals' is only available for these backends: [CPUTensorId]. when using pytorch built from master

Have you found a solution? I’m having the same problem