client-ocr/test-detection.html at master · siva-sub/client-ocr · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
<!DOCTYPE html>
<html>
<head>
  <title>OCR Detection Test</title>
  <script src="https://cdn.jsdelivr.net/npm/onnxruntime-web@1.22.0/dist/ort.min.js"></script>
</head>
<body>
  <h1>OCR Detection Test</h1>
  <input type="file" id="fileInput" accept="image/*">
  <canvas id="canvas" style="border: 1px solid black;"></canvas>
  <pre id="output"></pre>

  <script>
    const output = document.getElementById('output');

    function log(msg) {
      output.textContent += msg + '\n';
      console.log(msg);
    }

    document.getElementById('fileInput').addEventListener('change', async (e) => {
      const file = e.target.files[0];
      if (!file) return;

      const img = new Image();
      img.onload = async () => {
        const canvas = document.getElementById('canvas');
        const ctx = canvas.getContext('2d');
        canvas.width = img.width;
        canvas.height = img.height;
        ctx.drawImage(img, 0, 0);

        const imageData = ctx.getImageData(0, 0, img.width, img.height);

        try {
          // Initialize detection model
          const detModelUrl = 'https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/v3.3.0/onnx/PP-OCRv4/det/en_PP-OCRv3_det_infer.onnx';
          log('Loading detection model...');
          const session = await ort.InferenceSession.create(detModelUrl);
          log('Model loaded. Input names: ' + session.inputNames);
          log('Output names: ' + session.outputNames);

          // Preprocess image
          const { tensor, resizeInfo } = preprocessImage(imageData.data, img.width, img.height);

          // Run inference
          log('Running inference...');
          const results = await session.run({ x: tensor });
          log('Inference complete');

          // Get output
          const outputTensor = results[session.outputNames[0]];
          log('Output shape: ' + outputTensor.dims);
          log('Output type: ' + outputTensor.type);

          // Check output values
          const outputData = outputTensor.data;
          let min = Infinity, max = -Infinity, sum = 0;
          for (let i = 0; i < outputData.length; i++) {
            min = Math.min(min, outputData[i]);
            max = Math.max(max, outputData[i]);
            sum += outputData[i];
          }
          const avg = sum / outputData.length;
          log(`Output stats - Min: ${min}, Max: ${max}, Avg: ${avg}`);

          // Apply threshold and find boxes
          const thresh = 0.3;
          const boxThresh = 0.5;
          const bitmap = new Uint8Array(outputData.length);
          let countAboveThresh = 0;
          for (let i = 0; i < outputData.length; i++) {
            if (outputData[i] > thresh) {
              bitmap[i] = 255;
              countAboveThresh++;
            }
          }
          log(`Pixels above threshold (${thresh}): ${countAboveThresh} / ${outputData.length} (${(countAboveThresh/outputData.length*100).toFixed(2)}%)`);

          // Visualize detection map
          const [batch, channels, height, width] = outputTensor.dims;
          const detCanvas = document.createElement('canvas');
          detCanvas.width = width;
          detCanvas.height = height;
          const detCtx = detCanvas.getContext('2d');
          const detImageData = detCtx.createImageData(width, height);

          for (let y = 0; y < height; y++) {
            for (let x = 0; x < width; x++) {
              const idx = y * width + x;
              const val = outputData[idx] * 255;
              const pixelIdx = idx * 4;
              detImageData.data[pixelIdx] = val;
              detImageData.data[pixelIdx + 1] = val;
              detImageData.data[pixelIdx + 2] = val;
              detImageData.data[pixelIdx + 3] = 255;
            }
          }

          detCtx.putImageData(detImageData, 0, 0);
          document.body.appendChild(detCanvas);

        } catch (error) {
          log('Error: ' + error.message);
          console.error(error);
        }
      };

      img.src = URL.createObjectURL(file);
    });

    function preprocessImage(imageData, width, height) {
      // Resize to multiple of 32
      let ratio = 1.0;
      const maxSideLen = 960;

      if (Math.max(height, width) > maxSideLen) {
        ratio = height > width ? maxSideLen / height : maxSideLen / width;
      }

      let resizeH = Math.round(height * ratio);
      let resizeW = Math.round(width * ratio);

      resizeH = Math.round(resizeH / 32) * 32;
      resizeW = Math.round(resizeW / 32) * 32;

      resizeH = Math.max(resizeH, 32);
      resizeW = Math.max(resizeW, 32);

      log(`Resizing: ${width}x${height} → ${resizeW}x${resizeH}`);

      // Create resized image
      const canvas = new OffscreenCanvas(resizeW, resizeH);
      const ctx = canvas.getContext('2d');

      const inputImageData = new ImageData(new Uint8ClampedArray(imageData), width, height);
      const tempCanvas = new OffscreenCanvas(width, height);
      const tempCtx = tempCanvas.getContext('2d');
      tempCtx.putImageData(inputImageData, 0, 0);

      ctx.drawImage(tempCanvas, 0, 0, width, height, 0, 0, resizeW, resizeH);
      const resizedData = ctx.getImageData(0, 0, resizeW, resizeH).data;

      // Normalize
      const channels = 3;
      const normalized = new Float32Array(channels * resizeH * resizeW);

      // Use simple normalization for detection: img / 127.5 - 1.0
      for (let y = 0; y < resizeH; y++) {
        for (let x = 0; x < resizeW; x++) {
          const idx = (y * resizeW + x) * 4;
          const pixelIdx = y * resizeW + x;

          // Normalize to [-1, 1] range and rearrange to CHW format
          normalized[pixelIdx] = resizedData[idx] / 127.5 - 1.0;  // R
          normalized[resizeH * resizeW + pixelIdx] = resizedData[idx + 1] / 127.5 - 1.0;  // G
          normalized[2 * resizeH * resizeW + pixelIdx] = resizedData[idx + 2] / 127.5 - 1.0;  // B
        }
      }

      const tensor = new ort.Tensor('float32', normalized, [1, channels, resizeH, resizeW]);

      return {
        tensor,
        resizeInfo: {
          originalWidth: width,
          originalHeight: height,
          resizedWidth: resizeW,
          resizedHeight: resizeH,
          ratioW: resizeW / width,
          ratioH: resizeH / height
        }
      };
    }
  </script>
</body>
</html>