Update onnx.js

onnx.js

@@ -38,33 +38,40 @@ function ObjectDetection() {
   };
 
   const preprocessImage = async (imageSrc) => {
-    const img = new Image();
-    img.src = imageSrc;
-    await new Promise((resolve) => (img.onload = resolve));
+  const img = new Image();
+  img.src = imageSrc;
+  await new Promise((resolve) => (img.onload = resolve));
 
-    const canvas = document.createElement('canvas');
-    const context = canvas.getContext('2d');
+  const canvas = document.createElement('canvas');
+  const context = canvas.getContext('2d');
 
-    // Resize to model input size
-    const modelInputWidth = 300; // Replace with your model's input width
-    const modelInputHeight = 300; // Replace with your model's input height
-    canvas.width = modelInputWidth;
-    canvas.height = modelInputHeight;
+  // Resize to model input size
+  const modelInputWidth = 300; // Replace with your model's input width
+  const modelInputHeight = 300; // Replace with your model's input height
+  canvas.width = modelInputWidth;
+  canvas.height = modelInputHeight;
 
-    context.drawImage(img, 0, 0, modelInputWidth, modelInputHeight);
+  context.drawImage(img, 0, 0, modelInputWidth, modelInputHeight);
 
-    const imageData = context.getImageData(0, 0, modelInputWidth, modelInputHeight);
-    const inputTensor = new Float32Array(imageData.data.length / 4 * 3);
+  const imageData = context.getImageData(0, 0, modelInputWidth, modelInputHeight);
 
-    // Normalize pixel values and prepare input tensor
+  // Check the required data type
+  const isUint8 = true; // Set to true if your model expects uint8, false for float32
+
+  if (isUint8) {
+    // Create Uint8Array tensor
+    return new ort.Tensor('uint8', imageData.data, [1, modelInputHeight, modelInputWidth, 3]);
+  } else {
+    // Normalize to [0, 1] and create Float32Array tensor
+    const floatData = new Float32Array(imageData.data.length / 4 * 3);
     for (let i = 0, j = 0; i < imageData.data.length; i += 4) {
-      inputTensor[j++] = imageData.data[i] / 255;     // R
-      inputTensor[j++] = imageData.data[i + 1] / 255; // G
-      inputTensor[j++] = imageData.data[i + 2] / 255; // B
+      floatData[j++] = imageData.data[i] / 255;     // R
+      floatData[j++] = imageData.data[i + 1] / 255; // G
+      floatData[j++] = imageData.data[i + 2] / 255; // B
     }
-
-    return new ort.Tensor('float32', inputTensor, [1, 3, modelInputHeight, modelInputWidth]);
-  };
+    return new ort.Tensor('float32', floatData, [1, 3, modelInputHeight, modelInputWidth]);
+  }
+};
 
   const postprocessOutput = (output) => {
     const boxes = output['boxes'].data; // Replace 'boxes' with your model's output name