ROS2 run_slam example crash without error

[albertarla]

Hi, I was trying the sample of run_slam with ros2 foxy. To feed images I'm using the git repo mentioned on the tutorials (https://github.com/mirellameelo/dataset_publisher_ros2).
After entering at the subscription and feeding the image the code crashes without an error just after start global optimitzation module.

ros2 run stella_vslam_ros run_slam -v ~/workspace/orb_vocab.fbow -c ~/workspace/stella_vslam/example/aist/equirectangular.yaml  --debug
[2022-05-30 09:09:46.164] [D] CONSTRUCT: config
[2022-05-30 09:09:46.165] [I] config file loaded: /home/albert/workspace/stella_vslam/example/aist/equirectangular.yaml
[2022-05-30 09:09:46.165] [D] load camera model type
[2022-05-30 09:09:46.165] [D] load camera model parameters
[2022-05-30 09:09:46.165] [D] CONSTRUCT: camera::base
[2022-05-30 09:09:46.165] [D] CONSTRUCT: camera::equirectangular
[2022-05-30 09:09:46.165] [D] compute image bounds
[2022-05-30 09:09:46.165] [D] load ORB parameters
[2022-05-30 09:09:46.165] [D] CONSTRUCT: system
[2022-05-30 09:09:46.166] [I] 
original version of OpenVSLAM,
Copyright (C) 2019,
National Institute of Advanced Industrial Science and Technology (AIST)
All rights reserved.
stella_vslam (the changes after forking from OpenVSLAM),
Copyright (C) 2022, stella-cv, All rights reserved.

This is free software,
and you are welcome to redistribute it under certain conditions.
See the LICENSE file.

Camera:
  name: RICOH THETA S 960
  setup: monocular
  model: equirectangular
  fps: 30.0
  cols: 1920
  rows: 960
  color_order: RGB
Preprocessing:
  min_size: 800
  mask_rectangles:
    - [0.0, 1.0, 0.0, 0.1]
    - [0.0, 1.0, 0.84, 1.0]
    - [0.0, 0.2, 0.7, 1.0]
    - [0.8, 1.0, 0.7, 1.0]
Feature:
  name: default ORB feature extraction setting
  scale_factor: 1.2
  num_levels: 8
  ini_fast_threshold: 20
  min_fast_threshold: 7
Mapping:
  baseline_dist_thr_ratio: 0.02
  redundant_obs_ratio_thr: 0.95
LoopDetector:
  enabled: true
  reject_by_graph_distance: true
  min_distance_on_graph: 50

[2022-05-30 09:09:46.166] [I] loading ORB vocabulary: /home/workspace/orb_vocab.fbow
[2022-05-30 09:09:46.183] [D] CONSTRUCT: data::camera_database
[2022-05-30 09:09:46.183] [D] CONSTRUCT: data::map_database
[2022-05-30 09:09:46.183] [D] CONSTRUCT: data::bow_database
[2022-05-30 09:09:46.183] [D] CONSTRUCT: publish::frame_publisher
[2022-05-30 09:09:46.183] [D] CONSTRUCT: publish::map_publisher
[2022-05-30 09:09:46.184] [D] load maximum distance threshold where close keyframes could be found
[2022-05-30 09:09:46.184] [D] load maximum angle threshold between given pose and close keyframes
[2022-05-30 09:09:46.184] [D] CONSTRUCT: module::initializer
[2022-05-30 09:09:46.184] [D] CONSTRUCT: module::relocalizer
[2022-05-30 09:09:46.184] [D] CONSTRUCT: tracking_module
[2022-05-30 09:09:46.184] [D] CONSTRUCT: mapping_module
[2022-05-30 09:09:46.184] [D] load mapping parameters
[2022-05-30 09:09:46.184] [D] load monocular mappping parameters
[2022-05-30 09:09:46.184] [D] Use baseline_dist_thr_ratio: 0.02
[2022-05-30 09:09:46.184] [D] CONSTRUCT: loop_detector
[2022-05-30 09:09:46.184] [D] CONSTRUCT: global_optimization_module
[2022-05-30 09:09:46.184] [D] load depthmap factor
[2022-05-30 09:09:46.184] [D] CONSTRUCT: data::orb_params_database
[2022-05-30 09:09:46.200] [I] startup SLAM system
[2022-05-30 09:09:46.200] [I] start mapping module
[2022-05-30 09:09:46.200] [I] start global optimization module

It seems to subscribe to the topic and feed the image properly. I did visualize the frame reconverted to opencv matrix and it is working nice. Also, I tried the examples alone of stella_vslam (without ROS2) and it works as well. Anyone has the same problem?

Thanks in advance!

[ymd-stella]

After entering at the subscription and feeding the image the code crashes

How do you know it is after entering at the subscription and feeding the image?
How did you check?
Did you verify that the process does not exist?

[albertarla]

How do you know it is after entering at the subscription and feeding the image?

I edited the code to check if I can see the received image at system::create_monocular_frame()
Using cv::imshow() I receive the image properly and the conversion to greyscale is also correct.

Did you verify that the process does not exist?

What do you mean? Which process?

If it is working properly, it may be some of my dependencies... It's strange because the examples of stella_vslam are working properly and I don't receive any type of error or reason for the crash. If I run the code without publishing any image all works as it should, after the entering for the first time at the subscription dies. I'm with foxy and the las version of opencv.

[ymd-stella]

Describe the commands used for the build.
This may be the same problem described at xdspacelab/openvslam#299 (comment).
(edited: I noticed that you have confirmed that the images are received. This is not the case.)

By the way, MikiyaShibuya currently doesn't confirm the proper behavior of the ROS2 nodes which enable (only) SocketPublisher.
It seems that the SLAM node cannot subscribe a topic of image sequence.

[ymd-stella]

How about using docker?

try

$ docker build -t stella_vslam_ros2 -f Dockerfile.socket .
$ docker run --rm -it --name stella_vslam_ros2 --net=host -v $HOME:/host:rw stella_vslam_ros2 /bin/bash

For more information about SocketViewer, please refer to https://stella-cv.readthedocs.io/en/latest/docker.html#docker-image-of-server.

[albertarla]

With docker is working. Found the error, I have installed opencv when installing ROS2 and also for the vslam installation, it seems that it is causing some type of problem with the multiple installation.

Thanks for the support!