google / fluidnet Goto Github PK

the following command ends as per title line;

on this cloud I have neither wite permissions on /glob, nor sudo rights. I am trying a workaround using --local but it still fails, any ideas?

$ luarocks make tfluids-1-00.rockspec --local
...
make: *** [install] Error 1

Error: Failed installing dependency: https://raw.githubusercontent.com/torch/rocks/master/torch-scm-1.rockspec - Build error: Failed installing.
[u23885@c009 tfluids]$ luarocks make tfluids-1-00.rockspec

Error: Your user does not have write permissions in /glob/deep-learning/versions/torch/install/lib/luarocks/rocks
-- you may want to run as a privileged user or use your local tree with --local.

wget http://3d.csie.ntu.edu.tw/~dynamic/database/NTU3D.v1_0-999.zip
it seems this link no longer works.

Dear Sir/Madam,
I met a error when I run " luarocks make tfluids-1-00.rockspec". I dont know what the problem is. Output didnt show the reason. I need your help. Thanks.
Driver Version: 375.88 CUDA7.5/8 gcc4.8 ubuntu14.04 cmake3.2 Tesla K40
`-- Found Torch7 in /home/roseyu/su/distro/install
-- Compiling with OpenMP support
Compiling for CUDA architecture 3.5
Compiling with OpenGL support
Compiling with CUDA support.
-- Configuring done
-- Generating done
-- Build files have been written to: /home/roseyu/FliudNet/FluidNet/torch/tfluids/build
[ 33%] Building NVCC (Device) object CMakeFiles/tfluids.dir/generic/tfluids_generated_tfluids.cu.o
/home/roseyu/FliudNet/FluidNet/torch/tfluids/generic/tfluids.cu(1882): error: too few arguments in function call

1 error detected in the compilation of "/tmp/tmpxft_00008763_00000000-7_tfluids.cpp1.ii".
CMake Error at tfluids_generated_tfluids.cu.o.cmake:266 (message):
Error generating file
/home/roseyu/FliudNet/FluidNet/torch/tfluids/build/CMakeFiles/tfluids.dir/generic/./tfluids_generated_tfluids.cu.o

make[2]: *** [CMakeFiles/tfluids.dir/generic/tfluids_generated_tfluids.cu.o] Error 1
make[1]: *** [CMakeFiles/tfluids.dir/all] Error 2
make: *** [all] Error 2

Error: Build error: Failed building.`

when executing following:

./manta ../scenes/_trainingData.py --dim 3 --addModelGeometry True --addSphereGeometry True

the ouptut has entries such as:

FluidSolver::solvePressure skipping CorrectVelocity since res is nan!

According to #8 a possible workaround is turn down the gradient clipping magnitude first and see if that works. By default it's 1, but I would try as low as 0.2.

Any idea how to do that? Or any suggestions how to resolve this NaN problem?

Sorry to bother you.When run this command

luarocks make tfluids-1-00.rockspec

I only get the error FluidNet/torch/tfluids/generic/tfluids.cu(1882): error: too few arguments in function call.

So I find this line in file,like this

THCudaTensor_norm(state, tensor_pdelta_norm, tensor_pdelta, 2, 1);

Is there any change in function THCudaTensor_norm ?And I'am in trouble with finding the definition of THCudaTensor_norm.

Hi,

Is it possible to run the source code with Ubuntu 20.04.5 and CUDA 11? I cannot install torch7 successfully with my settings.

I would like to run FluidNet but i can't access this link(http://3d.csie.ntu.edu.tw/~dynamic/database/).

So how can i download the model files??

When I tried to #1. Generating the data, I got this:

When I tried to run 'fluid_net_train' I get:

qlua: lib/run_epoch.lua:221: criterion error is NaN or > 1e3.
stack traceback:
	[C]: at 0x7f551865f960
	[C]: in function 'error'
	lib/run_epoch.lua:221: in function 'opfunc'
	.../distro/install/share/lua/5.1/optim/adam.lua:38: in function 'optimMethod'
	lib/run_epoch.lua:320: in function 'runEpoch'
	fluid_net_train.lua:216: in main chunk

Looks like it is problem in training data but how can I check .bin files.
What your idea, what it could be?

Using Nvidia GTX 970 with 375.26 drivers and CUDA 8.0

From the documentation:

We assume that Torch7 is installed, otherwise follow the instructions here. We use the standard distro with the cuda SDK for cutorch and cunn and cudnn.

So the above instructions take me to http://torch.ch/ which I used to install torch as per http://torch.ch/docs/getting-started.html#_

But presumably the standard distro (= https://github.com/torch/distro ) and cudnn (= https://github.com/soumith/cudnn.torch ) are additional or different components.

Not knowing anything about torch I am fully in the dark as to which components should be installed. The documentation line I pasted above in italic is quite confusing

When running qlua fluid_net_train.lua -gpu 1 -dataset output_current_model_sphere -modelFilename myModel I get:

Try 'sleep --help' for more information.
sleep: invalid time interval ‘0,001’
Try 'sleep --help' for more information.
sleep: invalid time interval ‘0,001’========================>.]  319/320 
Try 'sleep --help' for more information.
sleep: invalid time interval ‘0,001’
Try 'sleep --help' for more information.
 [===========================================================>]  320/320 
sleep: invalid time interval ‘0,001’
Try 'sleep --help' for more information.
sleep: invalid time interval ‘0,001’
Try 'sleep --help' for more information.
==> Loaded 20480 samples
==> Creating model...
Number of input channels: 3
Model type: default
Bank 1:
Adding convolution: cudnn.SpatialConvolution(3 -> 16, 3x3, 1,1, 1,1)
Adding non-linearity: nn.ReLU (inplace true)
Bank 1:
Adding convolution: cudnn.SpatialConvolution(16 -> 16, 3x3, 1,1, 1,1)
Adding non-linearity: nn.ReLU (inplace true)
Bank 1:
Adding convolution: cudnn.SpatialConvolution(16 -> 16, 3x3, 1,1, 1,1)
Adding non-linearity: nn.ReLU (inplace true)
Bank 1:
Adding convolution: cudnn.SpatialConvolution(16 -> 16, 3x3, 1,1, 1,1)
Adding non-linearity: nn.ReLU (inplace true)
Adding convolution: cudnn.SpatialConvolution(16 -> 1, 1x1)
==> defining loss function
    using criterion nn.FluidCriterion: pLambda=0,00, uLambda=0,00, divLambda=1,00, borderWeight=1,0, borderWidth=3
==> Extracting model parameters
==> Defining Optimizer
    Using ADAM...
==> Profiling FPROP for 10 seconds with grid res 128
THCudaCheck FAIL file=/home/torstein/progs/FluidNet/torch/tfluids/generic/tfluids.cu line=119 error=8 : invalid device function
qlua: /home/torstein/torch/install/share/lua/5.1/tfluids/init.lua:516: cuda runtime error (8) : invalid device function at /home/torstein/progs/FluidNet/torch/tfluids/generic/tfluids.cu:119
stack traceback:
	[C]: at 0x7fdd9f648f50
	[C]: in function 'emptyDomain'
	/home/torstein/torch/install/share/lua/5.1/tfluids/init.lua:516: in function 'emptyDomain'
	fluid_net_train.lua:145: in main chunk

Using Nvidia GTX 770 with 367.57 drivers and 7.5.17 CUDA. Here's an overview over CUDA functions and required compute capability. The GPU in question has compute capability 3.0.

Here's the output from running './test.sh' in torch:
torch test.txt