kunle12 / dn_object_detect

Multiclass object detection ROS node based on YOLO/Darknet
BSD 2-Clause "Simplified" License
39 stars 24 forks source link

Getting stuck after "Loading the model data" with ROS Kinetic #5

Open Allopart opened 7 years ago

Allopart commented 7 years ago

I am trying to implement dn_object_detect on ROS Kinetic (Ubuntu 16.04, CUDA 8.0, OpenCV 2.4.13) but I keep running into the problem that it gets stuck after :

Loaded detection model data

The .cfg and .weight files are loaded correctly and the camera topic publishes values too. In the past I have successfully implemented it in ROS Indigo (Ubuntu 14.04, CUDA 7.5) and that is why I wanted to move further with it. Any suggestions? Kind regards

kunle12 commented 7 years ago

OK. It just happens that I will in fact test the code on this exact set up tomorrow (don't tell me that you are trying to run dn_object_detect on a TX2 board ;)). I will let you know how I go.

Allopart commented 7 years ago

Figured out the problem: for whatever reason Image_transport segfaults if using a compressed image. To solve it:

  1. In the .launch file, change "compressed" tag for "raw"
  2. In src/MultiClassObjectDetector.cpp, in line image_transport::TransportHints hints( "compressed" ) change to "raw"
kunle12 commented 7 years ago

Yes. This is what I have also encountered. There must be some bugs in the image transport code which is beyond the scope of this project. However, I have another code that is very similar to dn_object_detect, it works perfectly under the same environment. I'm struggling to find out what would possibly trigger this problem.