Resolved issues with multiple face detection

Author: esimov

examples/puploc/puploc.go

@@ -30,31 +30,30 @@ func FindFaces(pixels []uint8) uintptr {
 	dets := make([][]int, len(results))
 
 	for i := 0; i < len(results); i++ {
-		dets[i] = append(dets[i], results[i].Row, results[i].Col, results[i].Scale, int(results[i].Q))
-
+		dets[i] = append(dets[i], results[i].Row, results[i].Col, results[i].Scale, int(results[i].Q), 1)
 		// left eye
 		puploc := &pigo.Puploc{
 			Row:      results[i].Row - int(0.085*float32(results[i].Scale)),
 			Col:      results[i].Col - int(0.185*float32(results[i].Scale)),
 			Scale:    float32(results[i].Scale) * 0.4,
-			Perturbs: 100,
+			Perturbs: 50,
 		}
 		det := puplocClassifier.RunDetector(*puploc, *imageParams)
 		if det.Row > 0 && det.Col > 0 {
-			dets[i] = append(dets[i], det.Row, det.Col, int(det.Scale), int(results[i].Q))
+			dets[i] = append(dets[i], det.Row, det.Col, int(det.Scale), int(results[i].Q), 0)
 		}
 
 		// right eye
 		puploc = &pigo.Puploc{
 			Row:      results[i].Row - int(0.085*float32(results[i].Scale)),
 			Col:      results[i].Col + int(0.185*float32(results[i].Scale)),
 			Scale:    float32(results[i].Scale) * 0.4,
-			Perturbs: 100,
+			Perturbs: 50,
 		}
 
 		det = puplocClassifier.RunDetector(*puploc, *imageParams)
 		if det.Row > 0 && det.Col > 0 {
-			dets[i] = append(dets[i], det.Row, det.Col, int(det.Scale), int(results[i].Q))
+			dets[i] = append(dets[i], det.Row, det.Col, int(det.Scale), int(results[i].Q), 0)
 		}
 	}
 
@@ -68,7 +67,7 @@ func FindFaces(pixels []uint8) uintptr {
 
 		// Include as a first slice element the number of detected faces.
 		// We need to transfer this value in order to define the Python array buffer length.
-		coords = append([]int{len(dets), 0, 0, 0}, coords...)
+		coords = append([]int{len(dets), 0, 0, 0, 0}, coords...)
 
 		// Convert the slice into an array pointer.
 		s := *(*[]uint8)(unsafe.Pointer(&coords))
@@ -93,7 +92,7 @@ func clusterDetection(pixels []uint8, rows, cols int) []pigo.Detection {
 		Dim:    cols,
 	}
 	cParams := pigo.CascadeParams{
-		MinSize:     100,
+		MinSize:     60,
 		MaxSize:     600,
 		ShiftFactor: 0.1,
 		ScaleFactor: 1.1,
@@ -133,7 +132,7 @@ func clusterDetection(pixels []uint8, rows, cols int) []pigo.Detection {
 	dets := faceClassifier.RunCascade(cParams, 0.0)
 
 	// Calculate the intersection over union (IoU) of two clusters.
-	dets = faceClassifier.ClusterDetections(dets, 0.05)
+	dets = faceClassifier.ClusterDetections(dets, 0.0)
 
 	return dets
 }

examples/puploc/puploc.py

@@ -36,15 +36,15 @@ def process_frame(pixs):
 
 	if data_pointer :
 		buffarr = ((c_longlong * ARRAY_DIM) * MAX_NDETS).from_address(addressof(data_pointer.contents))
-		res = np.ndarray(buffer=buffarr, dtype=c_longlong, shape=(MAX_NDETS, 4,))
+		res = np.ndarray(buffer=buffarr, dtype=c_longlong, shape=(MAX_NDETS, 5,))
 
 		# The first value of the buffer aray represents the buffer length.
 		dets_len = res[0][0]
 		res = np.delete(res, 0, 0) # delete the first element from the array
 
 		# We have to consider the pupil pair added into the list.
 		# That's why we are multiplying the detection length with 3.
-		dets = list(res.reshape(-1, 4))[0:dets_len*3]
+		dets = list(res.reshape(-1, 5))[0:dets_len*3]
 		return dets
 
 # initialize the camera
@@ -69,21 +69,21 @@ def process_frame(pixs):
 		dets = process_frame(pixs) # pixs needs to be numpy.uint8 array
 
 		if dets is not None:
-			for det in dets[0:1]:
+			# We know that the detected faces are taking place in the first positions of the multidimensional array.
+			for det in dets:
 				if det[3] > 50:
-					cv2.circle(frame, (int(det[1]), int(det[0])), int(det[2]/2.0), (0, 0, 255), 2)
+					if det[4] == 1: # 1 == face; 0 == pupil
+						cv2.circle(frame, (int(det[1]), int(det[0])), int(det[2]/2.0), (0, 0, 255), 2)
+					else:
+						if showPupil:
+							cv2.circle(frame, (int(det[1]), int(det[0])), 4, (0, 0, 255), -1, 8, 0)
+						if showEyes:
+							cv2.rectangle(frame, 
+								(int(det[1])-int(det[2]), int(det[0])-int(det[2])), 
+								(int(det[1])+int(det[2]), int(det[0])+int(det[2])), 
+								(0, 255, 0), 2
+							)
 
-			for det in dets[1:]:
-				if det[3] > 50:
-					if showPupil:
-						cv2.circle(frame, (int(det[1]), int(det[0])), 4, (0, 0, 255), -1, 8, 0)
-					if showEyes:
-						cv2.rectangle(frame, 
-							(int(det[1])-int(det[2]), int(det[0])-int(det[2])), 
-							(int(det[1])+int(det[2]), int(det[0])+int(det[2])), 
-							(0, 255, 0), 2
-						)
-			
 	cv2.imshow('', frame)
 
 	key = cv2.waitKey(1)