client-ocr/test-detection-fixed.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
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
<!DOCTYPE html>
<html lang="en">
<head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <title>Test Detection - Fixed</title>
    <style>
        body {
            font-family: Arial, sans-serif;
            max-width: 1200px;
            margin: 0 auto;
            padding: 20px;
        }
        .container {
            display: flex;
            gap: 20px;
            margin-top: 20px;
        }
        .input-section, .output-section {
            flex: 1;
        }
        canvas {
            border: 1px solid #ccc;
            max-width: 100%;
            display: block;
            margin-top: 10px;
        }
        #log {
            background: #f5f5f5;
            padding: 10px;
            margin-top: 20px;
            height: 200px;
            overflow-y: auto;
            font-family: monospace;
            font-size: 12px;
        }
        .box-info {
            background: #e0f7fa;
            padding: 10px;
            margin: 10px 0;
            border-radius: 4px;
        }
    </style>
</head>
<body>
    <h1>Test Detection with PaddleOCR Normalization</h1>

    <div>
        <input type="file" id="fileInput" accept="image/*">
        <button id="detectBtn" disabled>Run Detection</button>
    </div>

    <div class="container">
        <div class="input-section">
            <h3>Input Image</h3>
            <canvas id="inputCanvas"></canvas>
        </div>
        <div class="output-section">
            <h3>Detection Result</h3>
            <canvas id="outputCanvas"></canvas>
            <div id="boxInfo"></div>
        </div>
    </div>

    <div id="log"></div>

    <script type="module">
        import { init as ortInit } from 'https://cdn.jsdelivr.net/npm/onnxruntime-web@1.16.3/dist/ort.min.js';

        let model = null;
        const log = document.getElementById('log');
        const fileInput = document.getElementById('fileInput');
        const detectBtn = document.getElementById('detectBtn');
        const inputCanvas = document.getElementById('inputCanvas');
        const outputCanvas = document.getElementById('outputCanvas');
        const boxInfo = document.getElementById('boxInfo');

        function logMessage(msg) {
            const timestamp = new Date().toISOString().slice(11, 23);
            log.innerHTML += `[${timestamp}] ${msg}\n`;
            log.scrollTop = log.scrollHeight;
        }

        // Initialize ONNX Runtime
        ortInit().then(ort => {
            window.ort = ort;
            logMessage('ONNX Runtime initialized');
            loadModel();
        });

        async function loadModel() {
            try {
                logMessage('Loading detection model...');
                const session = await ort.InferenceSession.create('/models/en-mobile/det.onnx');
                model = session;
                logMessage('Model loaded successfully');
                logMessage('Input names: ' + session.inputNames.join(', '));
                logMessage('Output names: ' + session.outputNames.join(', '));
                detectBtn.disabled = false;
            } catch (error) {
                logMessage('Error loading model: ' + error.message);
            }
        }

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

            const img = new Image();
            img.onload = () => {
                inputCanvas.width = img.width;
                inputCanvas.height = img.height;
                const ctx = inputCanvas.getContext('2d');
                ctx.drawImage(img, 0, 0);
                logMessage(`Image loaded: ${img.width}x${img.height}`);
            };
            img.src = URL.createObjectURL(file);
        });

        detectBtn.addEventListener('click', async () => {
            if (!model) {
                logMessage('Model not loaded');
                return;
            }

            const ctx = inputCanvas.getContext('2d');
            const imageData = ctx.getImageData(0, 0, inputCanvas.width, inputCanvas.height);

            try {
                logMessage('Starting detection...');
                const boxes = await runDetection(imageData);
                drawBoxes(boxes);
                displayBoxInfo(boxes);
            } catch (error) {
                logMessage('Detection error: ' + error.message);
                console.error(error);
            }
        });

        async function runDetection(imageData) {
            const { data, width, height } = imageData;

            // Resize configuration
            const config = {
                limit_side_len: 960,
                limit_type: 'max',
                mean: [0.485, 0.456, 0.406],
                std: [0.229, 0.224, 0.225],
                thresh: 0.3,
                box_thresh: 0.5,
                padding_vertical: 0.4,
                padding_horizontal: 0.6,
                minimum_area_threshold: 20
            };

            // Resize image
            const { resizedData, resizedWidth, resizedHeight, ratioH, ratioW } = resizeForDetection(
                data, width, height, config
            );

            // Preprocess
            const inputTensor = preprocessForDetection(resizedData, resizedWidth, resizedHeight, config);

            // Run inference
            const feeds = { [model.inputNames[0]]: inputTensor };
            const output = await model.run(feeds);

            // Post-process
            const boxes = postprocessDetection(output, resizedWidth, resizedHeight, ratioH, ratioW, width, height, config);

            return boxes;
        }

        function resizeForDetection(imageData, width, height, config) {
            let ratio = 1.0;

            if (config.limit_type === 'max') {
                if (Math.max(height, width) > config.limit_side_len) {
                    ratio = height > width
                        ? config.limit_side_len / height
                        : config.limit_side_len / width;
                }
            }

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

            // Make dimensions divisible by 32
            resizeH = Math.ceil(resizeH / 32) * 32;
            resizeW = Math.ceil(resizeW / 32) * 32;

            // Ensure minimum size
            resizeH = Math.max(resizeH, 32);
            resizeW = Math.max(resizeW, 32);

            logMessage(`Resize: ${width}x${height} → ${resizeW}x${resizeH} (ratio: ${ratio.toFixed(3)})`);

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

            // Create ImageData from input
            const inputImageData = new ImageData(
                new Uint8ClampedArray(imageData),
                width,
                height
            );

            // Create temporary canvas with original image
            const tempCanvas = new OffscreenCanvas(width, height);
            const tempCtx = tempCanvas.getContext('2d');
            tempCtx.putImageData(inputImageData, 0, 0);

            // Resize
            ctx.drawImage(tempCanvas, 0, 0, width, height, 0, 0, resizeW, resizeH);

            // Get resized data
            const resizedImageData = ctx.getImageData(0, 0, resizeW, resizeH);

            return {
                resizedData: resizedImageData.data,
                resizedWidth: resizeW,
                resizedHeight: resizeH,
                ratioH: resizeH / height,
                ratioW: resizeW / width
            };
        }

        function preprocessForDetection(imageData, width, height, config) {
            const channels = 3;
            const normalized = new Float32Array(channels * height * width);

            logMessage(`Preprocessing with mean=${config.mean}, std=${config.std}`);

            for (let y = 0; y < height; y++) {
                for (let x = 0; x < width; x++) {
                    const idx = (y * width + x) * 4;
                    const pixelIdx = y * width + x;

                    // PaddleOCR normalization: (pixel / 255.0 - mean) / std
                    normalized[pixelIdx] = (imageData[idx] / 255.0 - config.mean[0]) / config.std[0];  // R
                    normalized[height * width + pixelIdx] = (imageData[idx + 1] / 255.0 - config.mean[1]) / config.std[1];  // G
                    normalized[2 * height * width + pixelIdx] = (imageData[idx + 2] / 255.0 - config.mean[2]) / config.std[2];  // B
                }
            }

            return new ort.Tensor('float32', normalized, [1, channels, height, width]);
        }

        function postprocessDetection(output, resizedWidth, resizedHeight, ratioH, ratioW, originalWidth, originalHeight, config) {
            const outputName = Object.keys(output)[0];
            const outputTensor = output[outputName];
            const outputData = outputTensor.data;
            const shape = outputTensor.dims;

            logMessage(`Output shape: ${shape}`);

            let h, w;
            if (shape.length === 4) {
                [, , h, w] = shape;
            } else if (shape.length === 3) {
                [, h, w] = shape;
            } else {
                logMessage('Unexpected output shape');
                return [];
            }

            // Create binary map
            const bitmap = new Uint8Array(h * w);
            let pixelsAboveThreshold = 0;

            for (let y = 0; y < h; y++) {
                for (let x = 0; x < w; x++) {
                    const idx = y * w + x;
                    const prob = outputData[idx];
                    if (prob > config.thresh) {
                        bitmap[idx] = 255;
                        pixelsAboveThreshold++;
                    }
                }
            }

            logMessage(`Pixels above threshold: ${pixelsAboveThreshold} (${(pixelsAboveThreshold / bitmap.length * 100).toFixed(2)}%)`);

            // Find boxes
            const boxes = findTextBoxes(bitmap, w, h, resizedWidth, resizedHeight, config);

            // Convert back to original coordinates
            return boxes.map(box => ({
                x: Math.round(box.x / ratioW),
                y: Math.round(box.y / ratioH),
                width: Math.round(box.width / ratioW),
                height: Math.round(box.height / ratioH)
            }));
        }

        function findTextBoxes(bitmap, bitmapWidth, bitmapHeight, targetWidth, targetHeight, config) {
            const visited = new Uint8Array(bitmap.length);
            const boxes = [];

            for (let y = 0; y < bitmapHeight; y++) {
                for (let x = 0; x < bitmapWidth; x++) {
                    const idx = y * bitmapWidth + x;

                    if (bitmap[idx] === 255 && !visited[idx]) {
                        const component = findConnectedComponent(bitmap, visited, x, y, bitmapWidth, bitmapHeight);

                        const componentArea = (component.maxX - component.minX) * (component.maxY - component.minY);
                        if (componentArea > config.minimum_area_threshold) {
                            // Apply padding
                            const componentHeight = component.maxY - component.minY;
                            const vertPadding = Math.round(componentHeight * config.padding_vertical);
                            const horizPadding = Math.round(componentHeight * config.padding_horizontal);

                            let minX = Math.max(0, component.minX - horizPadding);
                            let maxX = Math.min(bitmapWidth - 1, component.maxX + horizPadding);
                            let minY = Math.max(0, component.minY - vertPadding);
                            let maxY = Math.min(bitmapHeight - 1, component.maxY + vertPadding);

                            // Scale to target dimensions
                            const scaleX = targetWidth / bitmapWidth;
                            const scaleY = targetHeight / bitmapHeight;

                            const box = {
                                x: Math.round(minX * scaleX),
                                y: Math.round(minY * scaleY),
                                width: Math.round((maxX - minX) * scaleX),
                                height: Math.round((maxY - minY) * scaleY)
                            };

                            if (box.width > 5 && box.height > 5) {
                                boxes.push(box);
                            }
                        }
                    }
                }
            }

            logMessage(`Found ${boxes.length} text boxes`);
            return boxes;
        }

        function findConnectedComponent(bitmap, visited, startX, startY, width, height) {
            const component = {
                minX: startX,
                maxX: startX,
                minY: startY,
                maxY: startY,
                pixels: 0
            };

            const stack = [[startX, startY]];

            while (stack.length > 0) {
                const [x, y] = stack.pop();
                const idx = y * width + x;

                if (x < 0 || x >= width || y < 0 || y >= height || visited[idx] || bitmap[idx] !== 255) {
                    continue;
                }

                visited[idx] = 1;
                component.pixels++;
                component.minX = Math.min(component.minX, x);
                component.maxX = Math.max(component.maxX, x);
                component.minY = Math.min(component.minY, y);
                component.maxY = Math.max(component.maxY, y);

                stack.push([x + 1, y], [x - 1, y], [x, y + 1], [x, y - 1]);
            }

            return component;
        }

        function drawBoxes(boxes) {
            outputCanvas.width = inputCanvas.width;
            outputCanvas.height = inputCanvas.height;

            const outCtx = outputCanvas.getContext('2d');
            const inCtx = inputCanvas.getContext('2d');

            // Copy original image
            outCtx.drawImage(inputCanvas, 0, 0);

            // Draw boxes
            outCtx.strokeStyle = 'red';
            outCtx.lineWidth = 2;

            boxes.forEach((box, i) => {
                outCtx.strokeRect(box.x, box.y, box.width, box.height);
                outCtx.fillStyle = 'red';
                outCtx.fillText(i.toString(), box.x + 5, box.y + 15);
            });
        }

        function displayBoxInfo(boxes) {
            boxInfo.innerHTML = `<h4>Detected ${boxes.length} boxes:</h4>`;
            boxes.forEach((box, i) => {
                boxInfo.innerHTML += `
                    <div class="box-info">
                        Box ${i}: x=${box.x}, y=${box.y}, w=${box.width}, h=${box.height}
                    </div>
                `;
            });
        }
    </script>
</body>
</html>