onnx-community

/

dfine_n_coco-ONNX

like 0

Follow

ONNX Community 1.22k

Thanks for your great work!

Unfortunately webgpu doesn't seem to work with this model.
Any help is appreciated, perhaps loading a different export from local file?

Here is the error I am getting:

"Error: using ceil() in shape computation is not yet supported for MaxPool".

Sample code:

import { pipeline } from 'https://cdn.jsdelivr.net/npm/@huggingface/[email protected]';
const IMG_URL = 'https://huggingface.co/datasets/Xenova/transformers.js-docs/resolve/main/cats.jpg';
const detector = await pipeline(
  'object-detection',
  'onnx-community/dfine_s_coco-ONNX',
  { device: 'webgpu' }
);

const img = new Image();
img.crossOrigin = 'anonymous';
img.src = IMG_URL;
await img.decode();

const canvas  = document.getElementById('canvas');
canvas.width  = img.naturalWidth;
canvas.height = img.naturalHeight;
const ctx     = canvas.getContext('2d');
ctx.drawImage(img, 0, 0);

const preds = await detector(IMG_URL, { threshold: 0.5 });

ctx.strokeStyle = '#00FF00';
ctx.lineWidth   = 2;
ctx.font        = '18px sans-serif';
ctx.fillStyle   = '#00FF00';

for (const { box, label, score } of preds) {
  const { xmin, ymin, xmax, ymax } = box;   // pixel coords
  ctx.strokeRect(xmin, ymin, xmax - xmin, ymax - ymin);
  ctx.fillText(`${label} ${(score * 100).toFixed(1)}%`,
               xmin + 4, ymin > 20 ? ymin - 6 : ymin + 18);
}