chrisgundling / yololight Goto Github PK

@chrisgundling Thank you for sharing your work. I am also very interested in a ROS implementation of Darkflow and I have several questions regarding the retraining with my own data.

Until now, I have been using Darkflow to train my own classes from custom data and config file to obtain custom checkpoint files. Can I use them with your ROS node as an input to detect object from a USB camera?

Also, what is the output (topic) of your node? Do you provide any launch file? And finally, is it compatible with Xenial/Kinetic?

Thank you.

Hi~ I have a little problems~
xxx is my own train model, and put it at cfg folder.
Then, when i execute the rosrun yolo_light darkflow.py, i encounter a problem:

How to fix this problem?
and if i want to use my own xx.weight and xx.data, where i can change path on put on?

Thanks~

When running the program, the image can only show where is the traffic light. Can I get the category of them?

Hello:)
I successfully installed and ran the object detection node but the video performance is quite low. Although the camera (ZED) streams 60fps (tried VGA and HD resolutions) the recognition window shows a video stream at a rate around 3-5 fps. Are there any hints to improve the performance?
My systems load (GPU and CPU) is around 30-45% (running TX2).

Thank you in advance!

May you kindly explain how to keep it working without using ROS Node ? I do not need it.

Hi! I am currently use this code with tiny-yolo-voc now, however I also want to compare with full-yolo-voc.

Since the darkflow should run on python3 but not python2, the ROS always runs on python2 makes me always occur error when I trying to load the full-yolo-voc.cfg structure.

Anyone have the solutions?