Expected isFloatingType(grads[i].type().scalarType())

[Gao-JT]

Dear @HaozheQi ,

Thanks for your excellent work ! Now I am trying to reproduce the results through the code you provided, but I got this error:

Traceback (most recent call last):
File "/home/gjt/.pycharm_helpers/pydev/pydevd.py", line 1668, in
main()
File "/home/gjt/.pycharm_helpers/pydev/pydevd.py", line 1662, in main
globals = debugger.run(setup['file'], None, None, is_module)
File "/home/gjt/.pycharm_helpers/pydev/pydevd.py", line 1072, in run
pydev_imports.execfile(file, globals, locals) # execute the script
File "/home/gjt/.pycharm_helpers/pydev/_pydev_imps/_pydev_execfile.py", line 18, in execfile
exec(compile(contents+"\n", file, 'exec'), glob, loc)
File "/home/gjt/P2B/train_tracking.py", line 165, in
loss.backward()
File "/home/gjt/anaconda3/envs/P2B/lib/python3.6/site-packages/torch/tensor.py", line 166, in backward
torch.autograd.backward(self, gradient, retain_graph, create_graph)
File "/home/gjt/anaconda3/envs/P2B/lib/python3.6/site-packages/torch/autograd/init.py", line 99, in backward
allow_unreachable=True) # allow_unreachable flag

RuntimeError: Expected isFloatingType(grads[i].type().scalarType()) to be true, but got false. (Could this error message be improved? If so, please report an enhancement request to PyTorch.)

My environment is:
python 3.6.9
pytorch 1.3.1
torchvision 0.4.2
cudatoolkit 10.0.30
cudnn 7.6.5
h5py 2.10.0
numpy 1.17.4
pprint 0.1
enum34 1.1.10
future 0.18.2
pandas 0.25.3
shapely 1.7b1
matplotlib 3.1.2
pomegranate 0.13.3
ipykernel 5.1.3.0
jupyter 1.0.0
imageio 2.6.1
pyquaternion 0.9.5

Do you know what's wrong with it? Do you know what's wrong with it? Looking forward to hearing from you. Thanks for your excellent work again !

[HaozheQi]

Dear @Gao-JT, I'm sorry I have no idea about this problem, you can keep this issue open to see if anyone else can give a solution.

[Gao-JT]

Dear @HaozheQi , I have solved this problem by following such method:

facebookresearch/votenet#72

[HaozheQi]

Great, and hope our work can give you some inspiration.

[CodeLHY]

Dear @HaozheQi , I have solved this problem by following such method:

facebookresearch/votenet#72

Hi, I have met the same problem, can you tell me how did you fixed this problem?

[cuge1995]

Dear @HaozheQi , I have solved this problem by following such method:
facebookresearch/votenet#72

Hi, I have met the same problem, can you tell me how did you fixed this problem?

Hello, I give you the solutions.
In pointnet2_utils.py

Line 308

inds = _ext.ball_query(new_xyz, xyz, radius, nsample)
ctx.mark_non_differentiable(inds)
return inds

replace return _ext.ball_query_score(new_xyz, xyz, score, radius, nsample)

And, Line 65

fps_inds = _ext.furthest_point_sampling(xyz, npoint)
ctx.mark_non_differentiable(fps_inds)
return fps_inds

replace return _ext.furthest_point_sampling(xyz, npoint)

[czy341181]

It works, thanks @cuge1995

[yancie-yjr]

#8 (comment)
hello, i follow your solution, and still fail

why do you use _ext.ball_query() to replace _ext.ball_query_score(), they are not the same function. can it change the final results?