search

kbatsos / CBMV

Welcome to the code repository of CBMV: A Coalesced Bidirectional Matching Volume for Disparity Estimation. CVPR 2018
GNU General Public License v3.0
53 stars 13 forks source link

cuda problem? #8

Open amaanda opened 5 years ago

amaanda commented 5 years ago

I am using a Ubuntu 18.04, CUDA 9.0, Python 2.7, Scikit-learn 0.17.1, Numpy 1.16.4, Scipy 1.2.2 and cPickle 1.71, opencv 3.2.0 and nvcc is 9.0, V9.0.176.

I am trying to run the following command python main.py --l ./datasets/ArtL/im0.png --r ./datasets/ArtL/im1.png --w 694 --h 554 --d 128 --model ./modelall_4m.rf --prob_save_path ./ But I get the following output:

Cost volume: ./.prob.npy not exists! Batch testing Iterations: 4.0 Progress |██████████████████████████████████████████████████| 100.0% Complete CUDA initialized. Error: no kernel image is available for execution on the device Error: no kernel image is available for execution on the device Error: no kernel image is available for execution on the device Error: no kernel image is available for execution on the device Error: no kernel image is available for execution on the device Error: no kernel image is available for execution on the device Error: no kernel image is available for execution on the device Error: no kernel image is available for execution on the device Error: no kernel image is available for execution on the device Error: no kernel image is available for execution on the device Error: no kernel image is available for execution on the device Error: no kernel image is available for execution on the device Error: no kernel image is available for execution on the device Error: no kernel image is available for execution on the device Error: no kernel image is available for execution on the device Error: no kernel image is available for execution on the device Error: no kernel image is available for execution on the device Error: no kernel image is available for execution on the device 0.346982002258 Saved ./results/disp.pfm

My guess is it's a problem with the CUDA architecture, any suggestion? Also, what does this output number mean? This is what my disp.pfm looks like:

image

Any help will be appreciated :) thanks!

amaanda commented 5 years ago

The error being printed in post.cu

kbatsos commented 5 years ago

Hi @amaanda this error is related to your GPU generation and the flag the code is compiled with. If you open the makefile in post you can see this flag "-arch=sm_52". Check the generation of your GPU and compile with the appropriate flag.

Let me know if this helps.

Without running the post-processing you can generate a disparity map from the raw cost.