Issue: RuntimeError: Not compiled with GPU support.

[tasuka98]

Hi, I tried running the script after installing all the dependencies and received the following error. I have installed both pytorch3d as well as detectron2 correctly by following the instruction for cuda 10.1 but I don't know why I am still receiving this error.

(base) tasuka@tasuka-TM1701:~/Desktop/meshrcnn$ python3 demo/demo.py --config-file configs/pix3d/meshrcnn_R50_FPN.yaml --input ./input/sofa.jpg --output output_demo --onlyhighest MODEL.WEIGHTS meshrcnn://meshrcnn_R50.pth

[05/04 01:44:37 demo]: Arguments: Namespace(config_file='configs/pix3d/meshrcnn_R50_FPN.yaml', focal_length=20.0, input='./input/sofa.jpg', onlyhighest=True, opts=['MODEL.WEIGHTS', 'meshrcnn://meshrcnn_R50.pth'], output='output_demo') /pytorch/torch/csrc/utils/python_arg_parser.cpp:756: UserWarning: This overload of nonzero is deprecated:
nonzero(Tensor input, *, Tensor out)
Consider using one of the following signatures instead:
nonzero(Tensor input, *, bool as_tuple)

/pytorch/aten/src/ATen/native/BinaryOps.cpp:81: UserWarning: Integer division of tensors using div or / is deprecated, and in a future release div will perform true division as in Python 3. Use true_divide or floor_divide (// in Python) instead.

Traceback (most recent call last):
File "demo/demo.py", line 227, in <module>
predictions = demo.run_on_image(img, focal_length=args.focal_length)

File "demo/demo.py", line 59, in run_on_image
predictions = self.predictor(image)

File "/home/tasuka/Desktop/detectron2/detectron2/engine/defaults.py", line 218, in __call__
predictions = self.model([inputs])[0]

File "/home/tasuka/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 550, in __call__
result = self.forward(*input, **kwargs)

File "/home/tasuka/Desktop/detectron2/detectron2/modeling/meta_arch/rcnn.py", line 108, in forward
return self.inference(batched_inputs)

File "/home/tasuka/Desktop/detectron2/detectron2/modeling/meta_arch/rcnn.py", line 170, in inference
results, _ = self.roi_heads(images, features, proposals, None)

File "/home/tasuka/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 550, in __call__
result = self.forward(*input, **kwargs)

File "/home/tasuka/Desktop/meshrcnn/meshrcnn/modeling/roi_heads/roi_heads.py", line 164, in forward
pred_instances = self.forward_with_given_boxes(features, pred_instances)

File "/home/tasuka/Desktop/meshrcnn/meshrcnn/modeling/roi_heads/roi_heads.py", line 185, in forward_with_given_boxes
instances = self._forward_shape(features, instances)

File "/home/tasuka/Desktop/meshrcnn/meshrcnn/modeling/roi_heads/roi_heads.py", line 357, in _forward_shape
pred_meshes = self.mesh_head(mesh_features, init_mesh)

File "/home/tasuka/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 550, in __call__
result = self.forward(*input, **kwargs)

File "/home/tasuka/Desktop/meshrcnn/meshrcnn/modeling/roi_heads/mesh_head.py", line 228, in forward
mesh, vert_feats = stage(x, mesh, vert_feats=vert_feats)

File "/home/tasuka/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 550, in __call__
result = self.forward(*input, **kwargs)

File "/home/tasuka/Desktop/meshrcnn/meshrcnn/modeling/roi_heads/mesh_head.py", line 183, in forward
vert_feats_nopos = F.relu(graph_conv(vert_feats, mesh.edges_packed()))

File "/home/tasuka/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 550, in __call__
result = self.forward(*input, **kwargs)

File "/home/tasuka/anaconda3/lib/python3.7/site-packages/pytorch3d/ops/graph_conv.py", line 71, in forward
neighbor_sums = gather_scatter(verts_w1, edges, self.directed)

File "/home/tasuka/anaconda3/lib/python3.7/site-packages/pytorch3d/ops/graph_conv.py", line 154, in forward
output = _C.gather_scatter(input, edges, directed, backward)

RuntimeError: Not compiled with GPU support.

[nitinsurya]

Tried fixing the pytorch version to 1.4 and torchvision to 0.5.0? This command fixed it for me:
conda install -c pytorch pytorch==1.4 torchvision==0.5.0
I suspect this error is raised now due to new pytorch version 1.5.

[gkioxari]

It seems that you haven't been able to build PyTorch3D with cuda. I am not an expert in resolving installation issues but could you take a look at PyTorch3D's issues under the installation label? I will also try to get some more insight into how we can resolve your issue.

[bottler]

@tasuka98 Can you paste the output of conda list please? Did you install pytorch3d with conda - which is the easiest way?

[tasuka98]

Tried fixing the pytorch version to 1.4 and torchvision to 0.5.0? This command fixed it for me:
conda install -c pytorch pytorch==1.4 torchvision==0.5.0
I suspect this error is raised now due to new pytorch version 1.5.

@nitinsurya I have tried to install them but the command only install them for CPU, and If I were to install them for cuda I need to downgrade it to 10.0, so should I do that instead?

[tasuka98]

Thanks everyone for their help!

@bottler I tried to reinstall pytorch3d along with torch and torchvision using conda and I was able to get the demo running by using the pre-built version of detectron2 for cuda10.1! Thank you.