deadprogram
commented
5 years ago Hello @zhlangsas you should know that as far as I know none of the OpenCV algorithms are re-entrant.
This applies to the CascadeClassifier. The solution is to create a new instance of the algorithm per goroutine. You would remove this code from your main function:
classifier := gocv.NewCascadeClassifier()
defer classifier.Close()
modelPath := "./haarcascade_frontalface_default.xml"
if !classifier.Load(modelPath) {
panic(fmt.Errorf("Error loading cascade file: %v", modelPath))
}Then replace your go routine code with something like this (untested) code:
for _, imgBytes := range imgs {
wg.Add(1)
go func(imgBytes []byte) {
classifier := gocv.NewCascadeClassifier()
defer classifier.Close()
modelPath := "./haarcascade_frontalface_default.xml"
if !classifier.Load(modelPath) {
panic(fmt.Errorf("Error loading cascade file: %v", modelPath))
}
img, _, err := image.Decode(bytes.NewReader(imgBytes))
if err != nil {
panic(err)
}
mat, err := gocv.ImageToMatRGB(img)
if err != nil {
panic(err)
}
rects := classifier.DetectMultiScale(mat)
fmt.Println(rects)
wg.Done()
}(imgBytes)
}Hope that helps!
zak905
DetectMultiScale panic when used concurrently
Description
When running cascade detection concurrently in goroutines I receive a variety of panics and segmentation faults.
Steps to Reproduce
Your Environment
env.shorenv.cmdscript before trying togo runorgo build? No