Issue with Preprocessing Input for TFLite Model [1, 640, 640, 3]

[yAlqubati]

Description

I’m trying to run an object detection TFLite model with the following specifications:

• Input shape: [1, 640, 640, 3]
• Output shape: [1, 25200, 7]

The main issue is preprocessing the input image to match the required format (float32). Below are the approaches I’ve tried so far, but none of them worked.

---

First Attempt

I used a straightforward method to read the image and convert it into float32:

import 'dart:io';
import 'package:tflite_flutter/tflite_flutter.dart';
import 'dart:developer' as developer;

class ScanController {
  void loadAndPredict() async {
    final interpreter = await Interpreter.fromAsset('assets/model.tflite');
    var image = File('assets/test.jpg');
    var imageBytes = await image.readAsBytes();

    // Convert image to float32
    final input = imageBytes.buffer.asFloat32List();

    // Expected output size: [1, 25200, 7]
    final output = List.filled(1 * 25200 * 7, 0.0).reshape([1, 25200, 7]);

    interpreter.run(input, output);
    developer.log(output.toString());
  }
}

Result: This approach failed because the image wasn’t resized, normalized, or correctly formatted for the model input.

---

Second Attempt

I used tflite_flutter_helper to preprocess the image, assuming it would handle resizing and normalization:

import 'dart:io';
import 'package:tflite_flutter/tflite_flutter.dart' as tflite;
import 'package:tflite_flutter_helper/tflite_flutter_helper.dart';
import 'dart:developer' as developer;

class ScanController {
  void loadAndPredict() async {
    try {
      final interpreter = await tflite.Interpreter.fromAsset('assets/model.tflite');
      developer.log('Model loaded successfully.');

      // Log input/output details
      var inputShape = interpreter.getInputTensor(0).shape;
      var inputType = interpreter.getInputTensor(0).type;
      var outputShape = interpreter.getOutputTensor(0).shape;
      developer.log('Input Shape: $inputShape, Type: $inputType');
      developer.log('Output Shape: $outputShape');

      var file = File('assets/test.jpg');

      // Preprocessing using TensorImage and ImageProcessor
      ImageProcessor imageProcessor = ImageProcessorBuilder()
          .add(ResizeOp(640, 640, ResizeMethod.BILINEAR))
          .add(NormalizeOp(0, 255))
          .build();

      TensorImage tensorImage = TensorImage.fromFile(file);
      tensorImage = imageProcessor.process(tensorImage);

      // Create output buffer
      TensorBuffer outputBuffer = TensorBuffer.createFixedSize(outputShape, tflite.TfLiteType.float32);

      // Run inference
      interpreter.run(tensorImage.buffer, outputBuffer.buffer);

      // Log the result
      var output = outputBuffer.getDoubleList();
      developer.log('Output: $output');
    } catch (e) {
      developer.log('Failed to load model: $e');
    }
  }
}

Result: This approach also failed. The main issue was with the tflite_flutter_helper package, which relies on outdated dependencies and is no longer actively maintained.

---

Request

Can someone please:

1. Explain the steps I need to follow to preprocess the input image to match the required format [1, 640, 640, 3] and type float32.
2. Provide guidance on resizing, normalizing, and converting the image effectively.
3. Suggest packages or dependencies to use, considering that tflite_flutter_helper is outdated.

Any help or example code would be greatly appreciated!