How to use the onnxjs.Tensor function in onnxjs

preprocess(data, width, height) {
      // data processing
      const dataTensor = ndarray(new Float32Array(data), [width, height, 4]);
      const dataProcessedTensor = ndarray(new Float32Array(width * height * 3), [1, 3, width, height]);

      ops.divseq(dataTensor, 128.0);
      ops.subseq(dataTensor, 1.0);

      ops.assign(dataProcessedTensor.pick(0, 0, null, null), dataTensor.pick(null, null, 2));
      ops.assign(dataProcessedTensor.pick(0, 1, null, null), dataTensor.pick(null, null, 1));
      ops.assign(dataProcessedTensor.pick(0, 2, null, null), dataTensor.pick(null, null, 0));

      const tensor = new onnx.Tensor(dataProcessedTensor.data, 'float32', [1, 3, width, height]);
      return tensor;
    }
}

async postprocess(tensor: Tensor, inferenceTime: number) {
    try {
      const originalOutput = new Tensor(tensor.data as Float32Array, 'float32', [1, 125, 13, 13]);
      const outputTensor = yoloTransforms.transpose(originalOutput, [0, 2, 3, 1]);

      // postprocessing
      const boxes = await yolo.postprocess(outputTensor, 20);
      boxes.forEach(box => {
        const {
          top, left, bottom, right, classProb, className,
        } = box;

        this.drawRect(left, top, right-left, bottom-top,
          `${className} Confidence: ${Math.round(classProb * 100)}% Time: ${inferenceTime.toFixed(1)}ms`);
      });
    } catch (e) {
      alert('Model is not valid!');
    }
  }

preprocess(ctx: CanvasRenderingContext2D): Tensor {
    const imageData = ctx.getImageData(0, 0, ctx.canvas.width, ctx.canvas.height);
    const { data, width, height } = imageData;
    // data processing
    const dataTensor = ndarray(new Float32Array(data), [width, height, 4]);
    const dataProcessedTensor = ndarray(new Float32Array(width * height * 3), [1, 3, width, height]);

    ops.assign(dataProcessedTensor.pick(0, 0, null, null), dataTensor.pick(null, null, 0));
    ops.assign(dataProcessedTensor.pick(0, 1, null, null), dataTensor.pick(null, null, 1));
    ops.assign(dataProcessedTensor.pick(0, 2, null, null), dataTensor.pick(null, null, 2));

    const tensor = new Tensor(new Float32Array(width* height* 3), 'float32', [1, 3, width, height]);
    (tensor.data as Float32Array).set(dataProcessedTensor.data);
    return tensor;
  }

const dataProcessedTensor = ndarray(new Float32Array(width * height * 3), [1, 3, width, height]);

    ops.assign(dataProcessedTensor.pick(0, 0, null, null), dataTensor.pick(null, null, 0));
    ops.assign(dataProcessedTensor.pick(0, 1, null, null), dataTensor.pick(null, null, 1));
    ops.assign(dataProcessedTensor.pick(0, 2, null, null), dataTensor.pick(null, null, 2));

    ops.divseq(dataProcessedTensor, 255);
    ops.subseq(dataProcessedTensor.pick(0, 0, null, null), 0.485);
    ops.subseq(dataProcessedTensor.pick(0, 1, null, null), 0.456);
    ops.subseq(dataProcessedTensor.pick(0, 2, null, null), 0.406);

    ops.divseq(dataProcessedTensor.pick(0, 0, null, null), 0.229);
    ops.divseq(dataProcessedTensor.pick(0, 1, null, null), 0.224);
    ops.divseq(dataProcessedTensor.pick(0, 2, null, null), 0.225);

    const tensor = new Tensor(new Float32Array(3 * width * height), 'float32', [1, 3, width, height]);
    (tensor.data as Float32Array).set(dataProcessedTensor.data);
    return tensor;
  }

const dataProcessedTensor = ndarray(new Float32Array(width * height * 3), [1, 3, width, height]);

    ops.assign(dataProcessedTensor.pick(0, 0, null, null), dataTensor.pick(null, null, 0));
    ops.assign(dataProcessedTensor.pick(0, 1, null, null), dataTensor.pick(null, null, 1));
    ops.assign(dataProcessedTensor.pick(0, 2, null, null), dataTensor.pick(null, null, 2));

    ops.divseq(dataProcessedTensor, 255);
    ops.subseq(dataProcessedTensor.pick(0, 0, null, null), 0.485);
    ops.subseq(dataProcessedTensor.pick(0, 1, null, null), 0.456);
    ops.subseq(dataProcessedTensor.pick(0, 2, null, null), 0.406);

    ops.divseq(dataProcessedTensor.pick(0, 0, null, null), 0.229);
    ops.divseq(dataProcessedTensor.pick(0, 1, null, null), 0.224);
    ops.divseq(dataProcessedTensor.pick(0, 2, null, null), 0.225);

    const tensor = new Tensor(new Float32Array(width * height * 3), 'float32', [1, 3, width, height]);
    (tensor.data as Float32Array).set(dataProcessedTensor.data);
    return tensor;
  }