Cannot run yolov4-tiny while yolov3-tiny works ok for jetson nano

[kyuhyong]

Describe the bug I was trying to detect an object with custom trained yolo weights. I trained weight using https://github.com/AlexeyAB/darknet on PC with GPU without any problem.

For this, I used my own data set "my_data" https://drive.google.com/file/d/1GcS9z2oqR1z5DuZpuAP-qEdJ7O8gntbx/view?usp=sharing

For yolov3-tiny

First I tried with yolov3-tiny by getting weight $ wget https://pjreddie.com/media/files/yolov3-tiny.weights then made pre-trained weight by $ ./darknet partial cfg/yolov3-tiny.cfg yolov3-tiny.weights yolov3-tiny.conv.15 15

Then train with following command $ ./darknet detector train my_data/detector.data my_data/cfg/yolov3-tiny_mydata.cfg yolov3-tiny.conv.15 -map >> log/my_tiny_data.log

For yolov4-tiny

Download weight from https://github.com/AlexeyAB/darknet/releases/download/darknet_yolo_v4_pre/yolov4-tiny.weights

Then made pre-trained wieght by $ ./darknet partial cfg/yolov4-tiny.cfg yolov4-tiny.weights yolov4-tiny.conv.15 15

I also tried with downloading yolov4-tiny.conv.29 from here $ wget https://github.com/AlexeyAB/darknet/releases/download/darknet_yolo_v4_pre/yolov4-tiny.conv.29

Train with

$ ./darknet detector train my_data/detector.data my_data/cfg/yolov4-tiny_cup.cfg yolov4-tiny.conv.29 -map >> log/my_tiny_data.log $ ./darknet detector train my_data/detector.data my_data/cfg/yolov4-tiny_cup.cfg yolov4-tiny.conv.15 -map >> log/my_tiny_data.log This is the result from training

I put both wegith, cfg file under yolo_network_config folder I made yaml file for both yolov3-tiny and yolov4-tiny

yaml file for yolov3-tiny

yolo_model:

config_file: name: yolov3-tiny_cup.cfg weight_file: name: yolov3-tiny_cup_final.weights threshold: value: 0.7 detection_classes: names: - cup

yaml file for yolov4-tiny

yolo_model:

config_file: name: yolov4-tiny_cup.cfg weight_file: name: yolov4-tiny_cup_final.weights threshold: value: 0.7 detection_classes: names: - cup

Screenshots While yolov3-tiny works just fine as screenshot below

my yolov4-tiny doesn't work with error like below.

Unused field: 'scale_x_y = 1.05'
Unused field: 'cls_normalizer = 1.0'
Unused field: 'iou_normalizer = 0.07'
Unused field: 'iou_loss = ciou'
Unused field: 'resize = 1.5'
Unused field: 'nms_kind = greedynms'
Unused field: 'beta_nms = 0.6'
   31 route  27
   32 conv    128  1 x 1 / 1    13 x  13 x 256   ->    13 x  13 x 128  0.011 BFLOPs
   33 upsample            2x    13 x  13 x 128   ->    26 x  26 x 128
   34 route  33 23
   35 conv    256  3 x 3 / 1    26 x  26 x 384   ->    26 x  26 x 256  1.196 BFLOPs
   36 conv     18  1 x 1 / 1    26 x  26 x 256   ->    26 x  26 x  18  0.006 BFLOPs
   37 yolo
Unused field: 'scale_x_y = 1.05'
Unused field: 'cls_normalizer = 1.0'
Unused field: 'iou_normalizer = 0.07'
Unused field: 'iou_loss = ciou'
Unused field: 'resize = 1.5'
Unused field: 'nms_kind = greedynms'
Unused field: 'beta_nms = 0.6'
Loading weights from /home/r1mini/catkin_ws/src/darknet_ros/darknet_ros/yolo_network_config/weights/yolov4-tiny_cup_final.weights...Done!
Waiting for image.
[ INFO:0] Initialize OpenCL runtime...
[darknet_ros-1] process has died [pid 12913, exit code -11, cmd /home/r1mini/catkin_ws/devel/lib/darknet_ros/darknet_ros camera/rgb/image_raw:=/main_camera/image_raw __name:=darknet_ros __log:=/home/r1mini/.ros/log/4191f940-f3f8-11eb-9662-38002506a2ca/darknet_ros-1.log].
log file: /home/r1mini/.ros/log/4191f940-f3f8-11eb-9662-38002506a2ca/darknet_ros-1*.log
all processes on machine have died, roslaunch will exit
shutting down processing monitor...
... shutting down processing monitor complete
done

What could have been wrong wit making weights? or trainig?

System (please complete the following information):

• OS: ubuntu 18.04 Jetpack 4.4

$ cat /etc/nv_tegra_release

R32 (release), REVISION: 4.4, GCID: 23942405, BOARD: t210ref, EABI: aarch64, DATE: Fri Oct 16 19:44:43 UTC 2020$ cat /etc/nv_tegra_release

R32 (release), REVISION: 4.4, GCID: 23942405, BOARD: t210ref, EABI: aarch64, DATE: Fri Oct 16 19:44:43 UTC 2020

• ROS version: melodic
• GPU when relevant: Jetson nano

[zeta0707]

Hi,

I also see same problem on Jetpack4.5.1.

OS: ubuntu 18.04 Jetpack 4.5.1

Thanks!

[DanielYG]

Hello,

Same problem here, any possible solution?

Thanks

[Ar-Ray-code]

This problem is similar to #328 .

[zeta0707]

@Ar-Ray-code Yes, I'm now using https://github.com/Tossy0423/yolov4-for-darknet_ros. It supports Yolo4, but opencv3.4.x is required.

Thanks!

[lawliet1997]

I am having hard time publishing the image topic after i run the launch file. What are you guys using to open up the camera? And how are you guys connecting your camera to Yolo? How did you get that image topic /darknet_ros/detection_image I am using Nvidia Jetson Xavier Raspberry Pie V2 Camera Ubuntu 18 with Ros Melodic