react-native-yolo 0.0.10 → 0.0.12

package/ios/HybridYoloModel.swift

@@ -32,8 +32,8 @@ public class HybridYoloModel: HybridYoloModelSpec {
             
             try localInterpreter.allocateTensors()
             
-            let inputTensor = try localInterpreter.inputTensor(at: 0)
-            let outputTensor = try localInterpreter.outputTensor(at: 0)
+            let inputTensor = try localInterpreter.input(at: 0)
+            let outputTensor = try localInterpreter.output(at: 0)
             
             let shape = inputTensor.shape.dimensions
             
@@ -81,7 +81,7 @@ public class HybridYoloModel: HybridYoloModelSpec {
         try localInterpreter.invoke()
         
         // Récupération des résultats du tenseur de sortie
-        let outputTensor = try localInterpreter.outputTensor(at: 0)
+        let outputTensor = try localInterpreter.output(at: 0)
         
         // Extraction et conversion de la structure des tenseurs [1, 300, 6] en Float
         let nativeOutputs = outputTensor.data.withUnsafeBytes { (ptr: UnsafeRawBufferPointer) -> [Float] in

package/ios/Loader/YoloModelLoader.swift

@@ -104,7 +104,7 @@ public class YoloModelLoader {
      */
     public func makeInputBuffer(interpreter: Interpreter) throws -> Data {
         // Récupérer le premier tenseur d'entrée du modèle TFLite
-        let inputTensor = try interpreter.inputTensor(at: 0)
+        let inputTensor = try interpreter.input(at: 0)
         let shape = inputTensor.shape.dimensions // Généralement: [1, 640, 640, 3]
         let dataType = inputTensor.dataType
         

package/ios/Utils/FrameJpegConverter.swift

@@ -1,41 +1,42 @@
 import Foundation
 import NitroModules
 import VisionCamera
+import CoreVideo
 
 public enum FrameJpegConverter {
     private static let tag = "YOLO_TAG_FrameJpegConverter"
-    
+
     public static func toJpegBytes(frame: any HybridFrameSpec, quality: Int = 80) -> [UInt8] {
-        // 1. Extraction et arrondi propre des dimensions du frame
-        let width = Int(frame.width.rounded())
-        let height = Int(frame.height.rounded())
-        
-        // 2. Conversion ultra-rapide via memcpy vers le format matériel NV12
-        let nv12 = Yuv420ToNv12Converter.convert(frame: frame, width: width, height: height)
-        
-        if nv12.isEmpty {
-            NSLog("[%@]: ❌ Failed to convert frame to NV12 array", tag)
-            return []
-        }
-        
-        // 3. Encodage matériel en JPEG via le GPU (CoreImage / CIContext)
-        let jpegBytes = Nv12JpegEncoder.encode(
-            nv12: nv12,
-            width: width,
-            height: height,
-            quality: quality
-        )
-        
-        if jpegBytes.isEmpty {
-            NSLog("[%@]: ❌ Failed to encode NV12 data to JPEG bytes", tag)
+        do {
+            let nativeBuffer = try frame.getNativeBuffer()
+
+            guard let rawPointer = UnsafeRawPointer(bitPattern: UInt(nativeBuffer.pointer)) else {
+                NSLog("[%@]: ❌ Failed to get native buffer pointer", tag)
+                return []
+            }
+
+            let pixelBuffer = Unmanaged<CVPixelBuffer>
+                .fromOpaque(rawPointer)
+                .takeUnretainedValue()
+
+            let jpegBytes = Nv12JpegEncoder.encode(
+                pixelBuffer: pixelBuffer,
+                quality: quality
+            )
+
+            if jpegBytes.isEmpty {
+                NSLog("[%@]: ❌ Failed to encode pixelBuffer to JPEG bytes", tag)
+                return []
+            }
+
+            return BitmapOrientationFixer.fix(
+                jpegBytes: jpegBytes,
+                frame: frame,
+                quality: quality
+            )
+        } catch {
+            NSLog("[%@]: ❌ Failed to convert frame to JPEG: %@", tag, "\(error)")
             return []
         }
-        
-        // 4. Redessin de sécurité via UIGraphicsImageRenderer pour fixer définitivement les pixels
-        return BitmapOrientationFixer.fix(
-            jpegBytes: jpegBytes,
-            frame: frame,
-            quality: quality
-        )
     }
-}
+}

package/ios/Utils/Nv12JpegEncoder.swift

@@ -1,58 +1,32 @@
 import Foundation
 import CoreImage
-import UIKit
+import CoreVideo
+import ImageIO
 
 public enum Nv12JpegEncoder {
-    // Réutiliser le CIContext permet d'éviter des fuites de mémoire massives à chaque frame
-    private static let ciContext = CIContext(options: [CIContextOption.useSoftwareRenderer: false])
-    
+    private static let ciContext = CIContext(options: [
+        CIContextOption.useSoftwareRenderer: false
+    ])
+
     public static func encode(
-        nv12: [UInt8],
-        width: Int,
-        height: Int,
+        pixelBuffer: CVPixelBuffer,
         quality: Int
     ) -> [UInt8] {
-        
-        // 1. Convertir la qualité (0-100 sur Android) en CGFloat (0.0-1.0 sur iOS)
         let compressionQuality = CGFloat(max(0, min(quality, 100))) / 100.0
-        
-        let ySize = width * height
-        let uvSize = ySize / 2
-        
-        // Sécurité : s'assurer que la taille du tableau correspond bien aux dimensions fournies
-        guard nv12.count >= (ySize + uvSize) else { return [] }
-        
-        // 2. Transformer le tableau d'octets [UInt8] en objet Data Swift
-        let rawData = Data(bytes: nv12, count: ySize + uvSize)
-        
-        // 3. Spécifier le format de couleur NV12 pour CoreImage (kCVPixelFormatType_420YpCbCr8BiPlanarVideoRange)
-        let imageOptions: [CIImageOption: Any] = [
-            .colorSpace: CGColorSpaceCreateDeviceRGB()
-        ]
-        
-        // On indique à CoreImage la structure exacte du NV12 (Plan 0: Y, Plan 1: UV entrelacé)
-        guard let ciImage = CIImage(
-            imageWithFormat: Int32(kCVPixelFormatType_420YpCbCr8BiPlanarVideoRange),
-            size: CGSize(width: width, height: height),
-            data: rawData,
-            rowBytes: width,
-            options: imageOptions
+
+        let ciImage = CIImage(cvPixelBuffer: pixelBuffer)
+        let colorSpace = CGColorSpaceCreateDeviceRGB()
+
+        guard let jpegData = ciContext.jpegRepresentation(
+            of: ciImage,
+            colorSpace: colorSpace,
+            options: [
+                kCGImageDestinationLossyCompressionQuality as CIImageRepresentationOption: compressionQuality
+            ]
         ) else {
             return []
         }
-        
-        // 4. Rendu de l'image GPU vers une structure d'image CoreGraphics
-        guard let cgImage = ciContext.createCGImage(ciImage, from: ciImage.extent) else {
-            return []
-        }
-        
-        // 5. Conversion en UIImage puis compression matérielle en JPEG
-        let uiImage = UIImage(cgImage: cgImage)
-        guard let jpegData = uiImage.jpegData(compressionQuality: compressionQuality) else {
-            return []
-        }
-        
-        // 6. Retourner le tableau d'octets natif [UInt8] requis par votre modèle
+
         return [UInt8](jpegData)
     }
-}
+}

package/ios/Utils/Yuv420ToNv12Converter.swift

@@ -8,7 +8,9 @@ public enum Yuv420ToNv12Converter {
     public static func convert(frame: any HybridFrameSpec, width: Int, height: Int) -> [UInt8] {
         do {
             let nativeBuffer = try frame.getNativeBuffer()
-            guard let rawPointer = nativeBuffer.pointer else { return [] }
+            guard let rawPointer = UnsafeRawPointer(bitPattern: UInt(nativeBuffer.pointer)) else {
+                return []
+            }
 
             let pixelBuffer = Unmanaged<CVPixelBuffer>
                 .fromOpaque(rawPointer)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "react-native-yolo",
-  "version": "0.0.10",
+  "version": "0.0.12",
   "description": "react-native-yolo is a react native package built with Nitro",
   "main": "./lib/commonjs/index.js",
   "module": "./lib/module/index.js",