search

princeton-vl / DPVO

Deep Patch Visual Odometry/SLAM
MIT License
670 stars 83 forks source link

Segmentation fault when run demo and some fix #9

Open dicarne opened 2 years ago

dicarne commented 2 years ago

OS: ubuntu 20.04LTS Gcc and g++: 10.3.0 Conda environment: same as environment.yml Cudnn: 8.3.2 Driver Version: 510.47.03 CUDA Version: 11.6

After installing the dependencies and downloading the model according to the README, run the README demo command, and it run failed. Without and error message.

$ CUDA_LAUNCH_BLOCKING=1 python demo.py --imagedir=movies/IMG_0494.MOV --calib=calib/iphone.txt --stride=5 --viz
Running with config...
BUFFER_SIZE: 2048
GRADIENT_BIAS: False
KEYFRAME_INDEX: 4
KEYFRAME_THRESH: 15.0
MIXED_PRECISION: True
MOTION_DAMPING: 0.5
MOTION_MODEL: DAMPED_LINEAR
OPTIMIZATION_WINDOW: 10
PATCHES_PER_FRAME: 96
PATCH_LIFETIME: 13
REMOVAL_WINDOW: 22
[1]    310720 segmentation fault (core dumped)  CUDA_LAUNCH_BLOCKING=1 python demo.py --imagedir=movies/IMG_0494.MOV   --viz

I found some method to fix it: in DPViewer/dpviewer/viewer.cpp image and image

Then I can run without error(sometimes). And I found the following change will be helpful, before the change, there was a certain probability of segmentation fault, after the change, no more segmentation fault. Presumably this is due to pointer initialization. image

Since I commented out the code to calculate the transformMatrix, the camera will not be updated. And I found that whenever I called the calculated code in the loop of this thread, it would definitely error out. Therefore, I create a member function specifically for calculating and updating the transformMatrix in main thread.

  1. New update transformMatrix function (Don't forget to write the function definition and mutex definition): image

  2. Lock when render thread draw points and poses: image

  3. Call it in python: image

Now, in the visualization view, the camera can render the pose normally.

But I observed that point cloud data was used, but my visualization view only had the camera pose and video, not any points.

I'm not very familiar with gl programming, so just a simple replace of the render point function. image

Finally, I can now run the visualization interface properly on my computer and update the poses and point cloud dynamically!

Here is all changes: https://github.com/dicarne/DPVO/commit/5f176844ecaf75a335c07ff667876de7907f3373 https://github.com/dicarne/DPVO/commit/677b0567d1b22470f6dcd7a63e03abdd49f3ab29

Some unanswered questions: Why, even with the same environment configuration as the author claims, I was unable to run the code of this project until I modified some of the code? I changed several computers and operating systems, and tried in docker, but all failed. I'm curious what the environment configuration is like to run it directly.

lahavlipson commented 2 years ago

We just created an official docker for DPVO with instructions on how to set it up: https://github.com/princeton-vl/DPVO_Docker

Hopefully this will be much easier; if you run into problems with the docker please follow up and I will try my best to help.

oscarfossey commented 2 years ago

Hello , thanks for the great repo! I have the same issue when using the docker file. Although i'm not sure it is the same causes because it refers to some XDG_RUNTIME_DIR.

(dpvo) root@laptop:/DPVO# python demo.py --imagedir=movies/IMG_0494.MOV --calib=calib/iphone.txt --stride=5 --viz 
Running with config...
BUFFER_SIZE: 2048
GRADIENT_BIAS: False
KEYFRAME_INDEX: 4
KEYFRAME_THRESH: 15.0
MIXED_PRECISION: True
MOTION_DAMPING: 0.5
MOTION_MODEL: DAMPED_LINEAR
OPTIMIZATION_WINDOW: 10
PATCHES_PER_FRAME: 96
PATCH_LIFETIME: 13
REMOVAL_WINDOW: 22
error: XDG_RUNTIME_DIR not set in the environment.
Segmentation fault (core dumped)
(dpvo) root@laptop:/DPVO#

EDIT: I change in the docker file the the repo link to the one from @dicarne and it worked.

https://github.com/dicarne/DPVO

Thank you very much both of you!

Launch-on-Titania commented 1 year ago

Hello , thanks for the great repo! I have the same issue when using the docker file. Although i'm not sure it is the same causes because it refers to some XDG_RUNTIME_DIR.

(dpvo) root@laptop:/DPVO# python demo.py --imagedir=movies/IMG_0494.MOV --calib=calib/iphone.txt --stride=5 --viz 
Running with config...
BUFFER_SIZE: 2048
GRADIENT_BIAS: False
KEYFRAME_INDEX: 4
KEYFRAME_THRESH: 15.0
MIXED_PRECISION: True
MOTION_DAMPING: 0.5
MOTION_MODEL: DAMPED_LINEAR
OPTIMIZATION_WINDOW: 10
PATCHES_PER_FRAME: 96
PATCH_LIFETIME: 13
REMOVAL_WINDOW: 22
error: XDG_RUNTIME_DIR not set in the environment.
Segmentation fault (core dumped)
(dpvo) root@laptop:/DPVO#

EDIT: I change in the docker file the the repo link to the one from @dicarne and it worked.

https://github.com/dicarne/DPVO

Thank you very much both of you!

Hi, I face the same issue while I use the docker env from the author. Have you solved it?