收藏
回答

如何解决InferenceSession推理报错CheckAndProcessReshape的问题?

模型:retinaface_mnet025_v2

下载地址:https://drive.google.com/file/d/1hzgOejAfCAB8WyfF24UkfiHD2FJbaCPi/view?usp=sharing (来自https://github.com/SthPhoenix/InsightFace-REST

报错:runInferenceSession:fail:run session fail(24:Failed in CheckAndProcessReshape() for Op[56]↵)

错误码:2004009

模型为float32,在onnxruntime-web上和python上都是可以正常运行的

const imageData = canvas.getImageData(0, 0, 640, 640)
const session = wx.createInferenceSession({
    model: modelPath,
    allowNPU: false,
    allowQuantize: false,
    precesionLevel: 4
  })
  await new Promise((resolve, reject) => {
    session.onError(reject)
    session.onLoad(resolve)
  })
    const data = new Float32Array(640 * 640 * 3)
    // NCHW
    const len = 640 * 640
    for (let i = 0; i < len; i++) {
      data[i] = imageData.data[i * 4]
      data[i + len] = imageData.data[i * 4 + 1]
      data[i + len * 2] = imageData.data[i * 4 + 2]
    }

    const results = await this.session.run({
      data: {
        type: 'float32',
        shape: [1, 3, 640, 640],
        data: data.buffer
      }
    })


回答关注问题邀请回答
收藏

1 个回答

  • Shirasawa
    Shirasawa
    03-09

    使用onnx-simplifier处理模型后直接报错:

    errMsg: "createInferenceSession:fail:create session fail(3:Failed to convert onnx model↵)"
    errno: 2004001
    
    03-09
    有用
    回复
登录 后发表内容