NMSBoxes fix & webcam example

Author: wimspaargaren

Makefile

@@ -15,7 +15,13 @@ lint:
 
 # Runs the examples
 bird-example:
-	cd cmd/example && go run .
+	@cd cmd/image && go run .
 
 street-example:
-	cd cmd/example && go run . -i ../../data/example_images/street.jpg
+	@cd cmd/image && go run . -i ../../data/example_images/street.jpg
+
+webcam-example:
+	@cd cmd/webcam && go run .
+
+cuda-example:
+	@cd cmd/cuda && go run .

README.md

@@ -1,16 +1,18 @@
 # Go YOLO V3
 
-This repository provides an implementation of the [Yolo V3](https://pjreddie.com/darknet/yolo/) object detection system in Go leveraging [gocv](https://github.com/hybridgroup/gocv).
+This repository provides a plug and play implementation of the [Yolo V3](https://pjreddie.com/darknet/yolo/) object detection system in Go, leveraging [gocv](https://github.com/hybridgroup/gocv).
 
-# Before you begin
+# Prerequisites
 
 Since this implementation builds on top of the [gocv](https://github.com/hybridgroup/gocv) library, make sure you either use one of the provided [docker images](https://github.com/hybridgroup/gocv/blob/release/Dockerfile) to run the example, or install the opencv dependencies on your system.
 
-Furthermore, make sure you've got the yolov3 models downloaded before running the examples. Simply run `$ make models`
+Furthermore, make sure you've got the yolov3 models downloaded before running the examples. 
 
-# Run the example
+Simply run `$ make models`
 
-Execute the bird example:
+# Run the examples
+
+## Bird example
 
 `$ make bird-example`
 
@@ -19,7 +21,7 @@ Output
 <img src="data/example_outputs/birds-output.png"
      alt="birds output"/>
 
-Execute the street example:
+## Street example
 
 `$ make street-example`
 
@@ -28,10 +30,33 @@ Output
 <img src="data/example_outputs/street-output.png"
      alt="street output"/>
 
+## Webcam example
+
+`$ make webcam-example`
+
+Note that this will not run smoothly on most machines, as the default net target type is set to `NetTargetCPU`. If you have cuda installed, adjust the net initialization to:
+```GOLANG
+	conf := yolov3.DefaultConfig()
+	// Adjust the backend and target type
+	conf.NetBackendType = gocv.NetBackendCUDA
+	conf.NetTargetType = gocv.NetTargetCUDA
+
+	// Create the net with created config
+	yolonet, err := yolov3.NewNetWithConfig(vehicleWeights, vehicleNetcfg, cocoNames, conf)
+	if err != nil {
+		log.WithError(err).Fatal("unable to create yolo net")
+	}
+```
+
+## Cuda example
+Execute 50 fps test render with cuda, also see the [CUDA](#CUDA) section.
+
+`$ make cuda-example`
+
 # CUDA
 
-If you're interested in running yolo in Go with CUDA support, check the `cmd/example_cuda` to see a dummy example and test results of running object detection at 50 fps.
+If you're interested in running yolo in Go with CUDA support, check the `cmd/example_cuda` to see a dummy example and test results of running object detection at 50 fps. The [gocv cuda README](https://github.com/hybridgroup/gocv/blob/release/cuda/README.md) provides detailed installation instructions.
 
 # Issues
 
-If you have any issues, feel free to open a PR or create an issue!
+If you have any issues, feel free to open a PR or create an issue!

cmd/example_cuda/README.md renamed to cmd/cuda/README.md

@@ -8,7 +8,7 @@ For CUDA installation instruction, see the gocv [CUDA README](https://github.com
 
 Software
 - Ubuntu 20.04 (64-bit)
-- Nvidea-smi 460.32.03
+- Nvidia-smi 460.32.03
 - CUDA version 11.2
 - OpenCV 4.5.1
 
@@ -26,4 +26,4 @@ Hardware
 
 <img src="nvidia-smi.png"
      alt="GPU Usage"
-     style="float: left; margin-right: 10px;" />
+     style="float: left; margin-right: 10px;" />

cmd/example_cuda/cpu-usage.png renamed to cmd/cuda/cpu-usage.png

@@ -0,0 +1,94 @@
+package main
+
+import (
+	"fmt"
+	"image"
+	"image/color"
+
+	log "github.com/sirupsen/logrus"
+	"github.com/wimspaargaren/yolov3"
+	"gocv.io/x/gocv"
+)
+
+const (
+	vehicleWeights = "../../data/yolov3/yolov3.weights"
+	vehicleNetcfg  = "../../data/yolov3/yolov3.cfg"
+	cocoNames      = "../../data/yolov3/coco.names"
+)
+
+func main() {
+	yolonet, err := yolov3.NewNet(vehicleWeights, vehicleNetcfg, cocoNames)
+	if err != nil {
+		log.WithError(err).Fatal("unable to create yolo net")
+	}
+
+	// Gracefully close the net when the program is done
+	defer func() {
+		err := yolonet.Close()
+		if err != nil {
+			log.WithError(err).Error("unable to gracefully close yolo net")
+		}
+	}()
+
+	videoCapture, err := gocv.OpenVideoCapture(0)
+	if err != nil {
+		log.WithError(err).Fatal("unable to start video capture")
+	}
+
+	window := gocv.NewWindow("Result Window")
+	defer func() {
+		err := window.Close()
+		if err != nil {
+			log.WithError(err).Error("unable to close window")
+		}
+	}()
+
+	frame := gocv.NewMat()
+	defer func() {
+		err := frame.Close()
+		if err != nil {
+			log.WithError(err).Errorf("unable to close image")
+		}
+	}()
+
+	for {
+		if ok := videoCapture.Read(&frame); !ok {
+			log.Error("unable to read videostram")
+		}
+		if frame.Empty() {
+			continue
+		}
+		detections, err := yolonet.GetDetections(frame)
+		if err != nil {
+			log.WithError(err).Fatal("unable to retrieve predictions")
+		}
+
+		drawDetections(&frame, detections)
+
+		window.IMShow(frame)
+		window.WaitKey(1)
+	}
+}
+
+func drawDetections(frame *gocv.Mat, detections []yolov3.ObjectDetection) {
+	for i := 0; i < len(detections); i++ {
+		detection := detections[i]
+		text := fmt.Sprintf("%s:%.2f", detection.ClassName, detection.Confidence)
+
+		// Create bounding box of object
+		blue := color.RGBA{0, 0, 255, 0}
+		gocv.Rectangle(frame, detection.BoundingBox, blue, 3)
+
+		// Add text background
+		black := color.RGBA{0, 0, 0, 0}
+		size := gocv.GetTextSize(text, gocv.FontHersheySimplex, 0.5, 1)
+		r := detection.BoundingBox
+		textBackground := image.Rect(r.Min.X, r.Min.Y-size.Y-1, r.Min.X+size.X, r.Min.Y)
+		gocv.Rectangle(frame, textBackground, black, int(gocv.Filled))
+
+		// Add text
+		pt := image.Pt(r.Min.X, r.Min.Y-4)
+		white := color.RGBA{255, 255, 255, 0}
+		gocv.PutText(frame, text, pt, gocv.FontHersheySimplex, 0.5, white, 1)
+	}
+}

cmd/example_cuda/main.go renamed to cmd/cuda/main.go

@@ -164,11 +164,19 @@ func (y *yoloNet) processOutputs(frame gocv.Mat, outputs []gocv.Mat) ([]ObjectDe
 			}
 		}
 	}
+	if len(bboxes) == 0 {
+		return detections, nil
+	}
 	indices := make([]int, len(bboxes))
 
 	gocv.NMSBoxes(bboxes, confidences, y.confidenceThreshold, y.nmsThreshold, indices)
 	result := []ObjectDetection{}
-	for _, indice := range indices {
+	for i, indice := range indices {
+		// If we encounter value 0 skip the detection
+		// except for the first indice
+		if i != 0 && indice == 0 {
+			continue
+		}
 		result = append(result, detections[indice])
 	}
 	return result, nil