search

Jumabek / darknet_scripts

Auxilary scripts to work with (YOLO) darknet deep learning famework. AKA -> How to generate YOLO anchors?
223 stars 96 forks source link
anchor-box anchors generated-anchors object-detection-anchor reproduce-yolov2-anchors visualize-genereted-anchors

readme

darknet_scripts - YOLO / YOLOv2 / YOLOv3 / YOLOv4 / YOLOv5 / YOLOv7 / YOLOv8

Auxilary scripts for working with darknet deep learning famework (2017)

  1. How to compute/reproduce YOLOv2 anchors for yolo-voc.cfg?
  2. How to visualize genereted anchors?
  3. Is gen_anchors.py same as YOLOv2 anchor computation?
  4. How to get anchors if My input for network is bigger than 416?
  5. How to plot YOLO loss
  6. YOLO and Anchors tutorial

How to compute/reproduce YOLOv2 anchors for yolo-voc.cfg?

  1. Download The Pascal VOC Data and unpack it to directory build\darknet\x64\data\voc will be created dir build\darknet\x64\data\voc\VOCdevkit\:

    1.1 Download file voc_label.py to dir build\darknet\x64\data\voc: http://pjreddie.com/media/files/voc_label.py

  2. Download and install Python for Windows: https://www.python.org/ftp/python/2.7.9/python-2.7.9rc1.amd64.msi

  3. Run command: python build\darknet\x64\data\voc\voc_label.py (to generate files: 2007_test.txt, 2007_train.txt, 2007_val.txt, 2012_train.txt, 2012_val.txt)

  4. Run command: type 2007_train.txt 2007_val.txt 2012_*.txt > train.txt

  5. Obtain anchors5.txt in generated_anchors/voc-reproduce folder by executing:

    python gen_anchors.py -filelist //path//to//voc//filelist/list//train.txt -output_dir generated_anchors/voc-reproduce -num_clusters 5

How to visualize genereted anchors?

After completing the steps above, execute 

python visualize_anchors.py -anchor_dir generated_anchors/voc-reproduce


Inside the generated_anchors/voc-reproduce directory you will have png visualization of the anchors

Is gen_anchors.py same as YOLOv2 anchor computation?

Yes, almost. Look at the two visualaziations below:


How to get anchors if My input for network is bigger than 416?

Simply change the lines here https://github.com/Jumabek/darknet_scripts/blob/master/gen_anchors.py#L17 to your input dimension. Then compute the anchors.

How to plot YOLO loss?

In order to plot a loss, you first need a log of the darknet train command For example,below command will save the log into log/aggregate-voc-tiny7.log 

darknet.exe detector train data/aggregate-voc-tiny7.data cfg/aggregate-voc-tiny7.cfg  backup/aggregate-voc-tiny7/aggregate-voc-tiny7_21000.weights >> log/aggregate-voc-tiny7.log -gpus 0,1

Once you have \path\to\log\aggregate-voc-tiny7.log, plot the loss by executing 

python plot_yolo_log.py \\path\\to\\log\\aggregate-voc-tiny7.log