The YOLOv4 APP can be applied to factory defect detection, medical image analysis, biological image analysis, industrial safety image analysis, mask image analysis, etc.
Russian Alexey Bochkovskiy, the maintainer of YOLO Darknet, discovered that the CSPNet detector developed by Wang Jianyao of the Institute of Information Technology of Academia Sinica and Director Liao Hongyuan was fast and good, so he invited Academia Sinica to develop YOLOv4 as a backbone. Various improvements have been made to the previous generation YOLOv3, which not only maintains a certain detection speed, but also greatly improves detection accuracy and reduces hardware usage.
[Operation steps and instructions]
Before using YOLOv4 training or object detection, please make sure that the dataset folder is selected correctly.
The "browse" button next to the "Select Dataset" field can open the data folder location for the user to confirm and modify.
Start preparing pre-processing, 1~5 of "Prepare" are the pre-processing part, respectively:
1.annotation pascal voc xml:Open the annotation webpage for image annotation.
2.convert yolo format：Convert the voc xml tag file into yolo format. Before pressing this button, please confirm whether the label.names category in the selected data set is filled in correctly.
3.prepare train txt: Generate train.txt.
4.prepare val txt: Generate valid.txt.
5.prepare config file: Open voc.data. If there is an increase or decrease in categories, please be sure to modify the number of classes. If there are 5 categories, fill in 5 (not counting background), and so on.
After finishing the pre-processing, you can start training. Training is divided into GPU and CPU. Please choose GPU or CPU mode according to your hardware.
There are several ways to infer, please choose GPU or CPU mode according to your hardware:
7.inference: inference a single image.
8.inference webcam: Inference webcam image.
9.inference folder: Inference all images in the test/images folder under the selected dataset folder.
10.inference api: Use the web page for image inference, press the “10.inference api” button, and then press the “11.inference api browser” button to open the web page.
99.edit yolov4.cfg: Edit the yolov4.cfg file. If there are any increase or decrease of categories, please modify the classes and filters in the three positions in yolov4.cfg.
99.browse data: Open the data folder.
99.auto labeling: Automatically label the images in the test folder, and output the labeling file to the auto_labeling folder.
99.calculate anchors: Use the training and verification data set to calculate the appropriate anchor points. If you want to modify the anchor points, please modify them in the three positions of yolov4.cfg.