모델 추론까지(안드환경)

packages/example/lib/vision_detector_views/face_detector_view.dart

@@ -3,6 +3,7 @@ import 'package:camera/camera.dart';
 import 'package:flutter/material.dart';
 import 'package:google_mlkit_face_detection/google_mlkit_face_detection.dart';
 import 'package:image/image.dart' as img;
+import 'package:image/image.dart';
 import 'package:tflite_flutter/tflite_flutter.dart';
 import 'detector_view.dart';
 import 'painters/face_detector_painter.dart';
@@ -88,14 +89,26 @@ class _FaceDetectorViewState extends State<FaceDetectorView> {
     // 사각형 영역을 구하기
     final squareRect = _getSquareRect(faces, inputImage.metadata!.size);
     if (squareRect != null) {
-      print("Updated squareRect: $squareRect");
-      image = cropImage(image, squareRect);
       if (image != null) {
-        // 예측 수행 부분 추가 가능
-        // _predict(image);
-        print(image);
+        //print('Cropped Image width: ${image.width}, height: ${image.height}');
+
+        final input = convertToFloat32List(await getCnnInput(image)).reshape([1,224,224,3]);
+
+        // print('input shape: ${input.shape}');
+        // print('input dtype: ${input.runtimeType}'); // TensorFlow의 float32인가 확인
+
+        // print(_interpreter.getInputTensors()[0]);
+        // print(_interpreter.getOutputTensors()[0]);
+        // Run inference and get the output
+        final output = List.filled(1 * 1, 0.0).reshape([1, 1]); // Assuming output size 1000
+        // print('output shape: ${output.shape}');
+        _interpreter.run(input, output);
+
+        setState(() {
+          _text = 'Inference Result: ${output[0]}';
+          print(_text);
+        });
       }
-
     } else {
       print("No square rect detected.");
     }
@@ -219,5 +232,67 @@ class _FaceDetectorViewState extends State<FaceDetectorView> {
       return null;
     }
   }
+  List<List<List<List<double>>>> _preprocessImage(img.Image image) {
+    // Step 1: Resize the image to 224x224
+    final img.Image resizedImage = img.copyResize(image, width: 224, height: 224);
+
+    // Step 2: Normalize the image pixel values to [0, 1]
+    final List<List<List<List<double>>>> input = List.generate(
+      1, // Batch size of 1
+      (_) => List.generate(
+        224, // Height
+        (y) => List.generate(
+          224, // Width
+          (x) {
+            // Extract pixel value (rgba format)
+            final pixel = resizedImage.getPixelSafe(x, y);
+
+            // Access RGBA components directly
+            final r = pixel.r / 255.0;
+            final g = pixel.g / 255.0;
+            final b = pixel.b / 255.0;
+
+            // Normalize to [-1, 1] if required
+            return [r * 2 - 1, g * 2 - 1, b * 2 - 1];
+          },
+        ),
+      ),
+    );
+
+    return input;
+  }
+
+
+
+
+  Float32List convertToFloat32List(List<List<List<double>>> input) {
+  final flattened = input.expand((row) => row.expand((col) => col)).toList();
+  return Float32List.fromList(flattened.map((e) => e.toDouble()).toList());
+}
+
+Future<List<List<List<double>>>> getCnnInput(img.Image image) async {
+  //print('cnn input');
+  // 이미지를 [224, 224] 크기로 리사이즈합니다.
+  img.Image resizedImage = img.copyResize(image, width: 224, height: 224);
+
+  // 결과를 저장할 배열을 초기화합니다.
+  List<List<List<double>>> result = List.generate(224, 
+    (_) => List.generate(224, 
+      (_) => List.filled(3, 0.0)));
+  // 픽셀 데이터를 [0, 1] 범위로 정규화하여 배열에 저장합니다.
+  for (int y = 0; y < resizedImage.height; y++) {
+    for (int x = 0; x < resizedImage.width; x++) {
+      Pixel pixel = resizedImage.getPixel(x, y);
+      double r = pixel[0] / 1;
+      double g = pixel[1] / 1;
+      double b = pixel[2] / 1;
+      result[y][x][0] = r; // Red
+      result[y][x][1] = g; // Green
+      result[y][x][2] = b; // Blue
+    }
+  }
+  //print(result.shape);
+  return result;
+}
 
 }

packages/example/pubspec.yaml

@@ -70,3 +70,4 @@ flutter:
   assets:
     - assets/ml/
     - assets/images/
+    - assets/food_resnet50.tflite