Skip to content

Commit

Permalink
removing scaling
Browse files Browse the repository at this point in the history
  • Loading branch information
@luicfrr
luicfrr committed Apr 16, 2024
1 parent 22c87fe commit 67d98f4
Show file tree
Hide file tree
Showing 3 changed files with 29 additions and 79 deletions.
2 changes: 1 addition & 1 deletion README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -149,7 +149,7 @@ export default function App() {
}
```

OBS: Returned face bounds, landmarks and contours are now natively downscaled to screen size. No extra function needed.
OBS: Returned face bounds are relative to frame coordinates and you need to scale them down to preview coordinates.

See example app for more details.

Expand Down
58 changes: 14 additions & 44 deletions android/src/main/java/com/visioncamerafacedetector/VisionCameraFaceDetectorPlugin.kt
View file
Original file line number Diff line number Diff line change
Expand Up @@ -92,27 +92,21 @@ class VisionCameraFaceDetectorPlugin(
}

private fun processBoundingBox(
boundingBox: Rect,
scaleX: Double,
scaleY: Double
boundingBox: Rect
): Map<String, Any> {
val bounds: MutableMap<String, Any> = HashMap()
val width = boundingBox.width().toDouble() * scaleX
val width = boundingBox.width().toDouble()

bounds["width"] = width
bounds["height"] = boundingBox.height().toDouble() * scaleY
bounds["x"] = windowWidth - (width + (
boundingBox.left.toDouble() * scaleX
))
bounds["y"] = boundingBox.top.toDouble() * scaleY
bounds["height"] = boundingBox.height().toDouble()
bounds["x"] = windowWidth - (width + boundingBox.left.toDouble())
bounds["y"] = boundingBox.top.toDouble()

return bounds
}

private fun processLandmarks(
face: Face,
scaleX: Double,
scaleY: Double
face: Face
): Map<String, Any> {
val faceLandmarksTypes = intArrayOf(
FaceLandmark.LEFT_CHEEK,
Expand Down Expand Up @@ -155,18 +149,16 @@ class VisionCameraFaceDetectorPlugin(
}
val point = landmark.position
val currentPointsMap: MutableMap<String, Double> = HashMap()
currentPointsMap["x"] = point.x.toDouble() * scaleX
currentPointsMap["y"] = point.y.toDouble() * scaleY
currentPointsMap["x"] = point.x.toDouble()
currentPointsMap["y"] = point.y.toDouble()
faceLandmarksTypesMap[landmarkName] = currentPointsMap
}

return faceLandmarksTypesMap
}

private fun processFaceContours(
face: Face,
scaleX: Double,
scaleY: Double
face: Face
): Map<String, Any> {
val faceContoursTypes = intArrayOf(
FaceContour.FACE,
Expand Down Expand Up @@ -221,8 +213,8 @@ class VisionCameraFaceDetectorPlugin(
val pointsMap: MutableMap<String, Map<String, Double>> = HashMap()
for (j in points.indices) {
val currentPointsMap: MutableMap<String, Double> = HashMap()
currentPointsMap["x"] = points[j].x.toDouble() * scaleX
currentPointsMap["y"] = points[j].y.toDouble() * scaleY
currentPointsMap["x"] = points[j].x.toDouble()
currentPointsMap["y"] = points[j].y.toDouble()
pointsMap[j.toString()] = currentPointsMap
}

Expand Down Expand Up @@ -251,16 +243,6 @@ class VisionCameraFaceDetectorPlugin(

val rotation = orientation!!.toDegrees()
val image = InputImage.fromMediaImage(frameImage!!, rotation)
val scaleX: Double
val scaleY: Double
if (rotation == 270 || rotation == 90) {
scaleX = windowWidth.toDouble() / image.height
scaleY = windowHeight.toDouble() / image.width
} else {
scaleX = windowWidth.toDouble() / image.width
scaleY = windowHeight.toDouble() / image.height
}

val task = faceDetector!!.process(image)
val faces = Tasks.await(task)
val facesList = ArrayList<Map<String, Any?>>()
Expand All @@ -269,11 +251,7 @@ class VisionCameraFaceDetectorPlugin(
val map: MutableMap<String, Any?> = HashMap()

if (runLandmarks) {
map["landmarks"] = processLandmarks(
face,
scaleX,
scaleY
)
map["landmarks"] = processLandmarks(face)
}

if (runClassifications) {
Expand All @@ -283,11 +261,7 @@ class VisionCameraFaceDetectorPlugin(
}

if (runContours) {
map["contours"] = processFaceContours(
face,
scaleX,
scaleY
)
map["contours"] = processFaceContours(face)
}

if (trackingEnabled) {
Expand All @@ -297,11 +271,7 @@ class VisionCameraFaceDetectorPlugin(
map["rollAngle"] = face.headEulerAngleZ.toDouble()
map["pitchAngle"] = face.headEulerAngleX.toDouble()
map["yawAngle"] = face.headEulerAngleY.toDouble()
map["bounds"] = processBoundingBox(
face.boundingBox,
scaleX,
scaleY
)
map["bounds"] = processBoundingBox(face.boundingBox)
facesList.add(map)
}

Expand Down
48 changes: 14 additions & 34 deletions ios/VisionCameraFaceDetector.swift
View file
Original file line number Diff line number Diff line change
Expand Up @@ -92,24 +92,20 @@ public class VisionCameraFaceDetector: FrameProcessorPlugin {
}

func processBoundingBox(
from face: Face,
scaleX: CGFloat,
scaleY: CGFloat
from face: Face
) -> [String:Any] {
let boundingBox = face.frame

return [
"width": boundingBox.width * scaleX,
"height": boundingBox.height * scaleY,
"x": boundingBox.origin.x * scaleX,
"y": boundingBox.origin.y * scaleY
"width": boundingBox.width,
"height": boundingBox.height,
"x": boundingBox.origin.x,
"y": boundingBox.origin.y
]
}

func processLandmarks(
from face: Face,
scaleX: CGFloat,
scaleY: CGFloat
from face: Face
) -> [String:[String: CGFloat?]] {
let faceLandmarkTypes = [
FaceLandmarkType.leftCheek,
Expand Down Expand Up @@ -141,8 +137,8 @@ public class VisionCameraFaceDetector: FrameProcessorPlugin {
for i in 0..<faceLandmarkTypes.count {
let landmark = face.landmark(ofType: faceLandmarkTypes[i]);
let position = [
"x": landmark?.position.x ?? 0.0 * scaleX,
"y": landmark?.position.y ?? 0.0 * scaleY
"x": landmark?.position.x ?? 0.0,
"y": landmark?.position.y ?? 0.0
]
faceLandMarksTypesMap[faceLandmarksTypesStrings[i]] = position
}
Expand All @@ -151,9 +147,7 @@ public class VisionCameraFaceDetector: FrameProcessorPlugin {
}

func processFaceContours(
from face: Face,
scaleX: CGFloat,
scaleY: CGFloat
from face: Face
) -> [String:[[String:CGFloat]]] {
let faceContoursTypes = [
FaceContourType.face,
Expand Down Expand Up @@ -199,8 +193,8 @@ public class VisionCameraFaceDetector: FrameProcessorPlugin {
if let points = contour?.points {
for point in points {
let currentPointsMap = [
"x": point.x * scaleX,
"y": point.y * scaleY,
"x": point.x,
"y": point.y
]

pointsArray.append(currentPointsMap)
Expand Down Expand Up @@ -242,19 +236,13 @@ public class VisionCameraFaceDetector: FrameProcessorPlugin {
let image = VisionImage(buffer: frame.buffer)
image.orientation = .up

let scaleX = screenBounds.size.width / CGFloat(frame.width)
let scaleY = screenBounds.size.height / CGFloat(frame.height)
var faceList: [Any] = []
let faces: [Face] = try faceDetector!.results(in: image)
for face in faces {
var map: [String: Any] = [:]

if runLandmarks {
map["landmarks"] = processLandmarks(
from: face,
scaleX: scaleX,
scaleY: scaleY
)
map["landmarks"] = processLandmarks(from: face)
}

if runClassifications {
Expand All @@ -264,11 +252,7 @@ public class VisionCameraFaceDetector: FrameProcessorPlugin {
}

if runContours {
map["contours"] = processFaceContours(
from: face,
scaleX: scaleX,
scaleY: scaleY
)
map["contours"] = processFaceContours(from: face)
}

if trackingEnabled {
Expand All @@ -278,11 +262,7 @@ public class VisionCameraFaceDetector: FrameProcessorPlugin {
map["rollAngle"] = face.headEulerAngleZ
map["pitchAngle"] = face.headEulerAngleX
map["yawAngle"] = face.headEulerAngleY
map["bounds"] = processBoundingBox(
from: face,
scaleX: scaleX,
scaleY: scaleY
)
map["bounds"] = processBoundingBox(from: face)

faceList.append(map)
}
Expand Down

0 comments on commit 67d98f4

Please sign in to comment.