Spaces:
Running
Running
File size: 2,980 Bytes
c89e680 | 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 | /* Decode standard YOLOv8 exports: batch 1, xywh + nine class scores, no NMS. */
(function (scope) {
'use strict';
function iou(a, b) {
const width = Math.max(0, Math.min(a[2], b[2]) - Math.max(a[0], b[0]));
const height = Math.max(0, Math.min(a[3], b[3]) - Math.max(a[1], b[1]));
const overlap = width * height;
const areaA = Math.max(0, a[2] - a[0]) * Math.max(0, a[3] - a[1]);
const areaB = Math.max(0, b[2] - b[0]) * Math.max(0, b[3] - b[1]);
return overlap / Math.max(1e-12, areaA + areaB - overlap);
}
function decode(output, names, transform, confidence, threshold = 0.7) {
const dims = output.dims;
const channels = 4 + names.length;
if (dims.length !== 3 || dims[0] !== 1 ||
(dims[1] !== channels && dims[2] !== channels)) {
throw new Error('Unsupported ONNX output. Export YOLOv8 with nine classes and nms=False.');
}
const channelFirst = dims[1] === channels;
const count = channelFirst ? dims[2] : dims[1];
const value = (candidate, channel) => output.data[
channelFirst ? channel * count + candidate : candidate * channels + channel
];
const candidates = [];
for (let index = 0; index < count; index += 1) {
let classId = 0;
let score = -Infinity;
for (let cls = 0; cls < names.length; cls += 1) {
const current = value(index, cls + 4);
if (current > score) { score = current; classId = cls; }
}
if (!Number.isFinite(score) || score < confidence) continue;
const cx = value(index, 0), cy = value(index, 1);
const width = value(index, 2), height = value(index, 3);
if (![cx, cy, width, height].every(Number.isFinite) || width <= 0 || height <= 0) continue;
candidates.push({
raw: [cx - width / 2, cy - height / 2, cx + width / 2, cy + height / 2],
confidence: score, class_id: classId, class_name: names[classId],
});
}
candidates.sort((a, b) => b.confidence - a.confidence);
// Match Ultralytics' pre-NMS candidate cap; class-aware NMS, max_det=100.
const retained = [];
for (const candidate of candidates.slice(0, 30000)) {
if (retained.some(item => item.class_id === candidate.class_id && iou(item.raw, candidate.raw) > threshold)) continue;
retained.push(candidate);
if (retained.length === 100) break;
}
const clamp = (value, maximum) => Math.max(0, Math.min(maximum, value));
return retained.map(item => ({
class_id: item.class_id,
class_name: item.class_name,
confidence: item.confidence,
box: item.raw.map((value, axis) => {
const horizontal = axis % 2 === 0;
const coordinate = (value - (horizontal ? transform.left : transform.top)) / transform.ratio;
return Math.round(clamp(coordinate, horizontal ? transform.width : transform.height) * 100) / 100;
}),
}));
}
scope.PaddyDetectionCore = { decode, iou };
})(typeof self !== 'undefined' ? self : globalThis);
|