Darknet: About calculate recall problem?

[kongjibai]

Hello! I have trained yolov3 model use the VOC dataset as mentioned pjreddie.com/darknet/yolo I have two questions as follow,

1. what are the mean of 2nd col and 3rd col which I labeled?
2. When I use the command ./darknet detector recall cfg/voc.data cfg/yolov3-voc.cfg backup/yolov3-voc_40000.weights calculate the test dataset recall, there is such result. I want to know why appear nan% of IOU suddenly, and this also lead to vary low recall value.
   
   

[patilameya825]

Hi! For nan in recall script of detector.c in yolo, refer 952 Issue. This helped me to solve the same error which I faced.

[zhaonann]

I just want to know how can you run the command,whether change some path for test pictures.
I try it but something is wrong.
Thanks!!!

[nuannuan1991]

I have the same problem,
how to modify validate_detector_recall function?

[jaelim]

@kongjibai What changes have you made before running your command to validate your trained? I used exactly the same command, but I get this error output:

Even though I had the correct file path to my validation test.txt in the .data configuration file.

[fate3439]

@jaelim @kongjibai
I encountered the exact same error as jaelim. darknet detector train/test/valid all works fine but the recall does not.

[fate3439]

Hi Guys, I found a walk around. Simply clone this repo https://github.com/AlexeyAB/darknet
Compile this darknet as usual. Now you can use ./darknet detector valid/recall/map, with the last one giving you mean Average Precision.

[jaelim]

@fate3439 I found out that I needed to modified the examples/detector.c and re-compile again. Then it worked.

[srhtyldz]

Hi Guys, I found a walk around. Simply clone this repo https://github.com/AlexeyAB/darknet
Compile this darknet as usual. Now you can use ./darknet detector valid/recall/map, with the last one giving you mean Average Precision.

What is the difference between valid and mAP ?