CODEWITHSUNDEEP
react-nativetensorflowtensorflow.js
nixoschu
nixoschu

Reputation: 544

Error: The shape of dict['ToFloat'] provided in model.execute(dict) must be [-1,-1,-1,3], but was [200,150,3]

Currently building a ReactNative App to detect Objects using the TensorFlow React Native Library.

Not having any experience with TensorFlow, I followed the examples and Guides, setting up the Expo Camera with the HOC of tfjs-reactnative. (https://js.tensorflow.org/api_react_native/latest/#cameraWithTensors)

         <TensorCamera
          // Standard Camera props
          style={styles.camera}
          type={Camera.Constants.Type.back}
          // Tensor related props
          cameraTextureHeight={textureDims.height}
          cameraTextureWidth={textureDims.width}
          resizeHeight={200}
          resizeWidth={150}
          resizeDepth={3}
          onReady={handleCameraStream}
          autorender={true}
        />

I check if TensorFlow is ready and load my Model from a Webserver

        await tf.ready();
        const model = await tf.loadGraphModel(
          'https://MODEL-DOMAIN.com/model.json'
        );

Camera is working and calling the handling function as a callback.

  const handleCameraStream = (images, updatePreview, gl) => {
    const loop = async () => {
      const nextImageTensor = images.next().value;

        if (detectionModel) {
          try {
            console.log(nextImageTensor);
            const prediction = await detectionModel.predict(nextImageTensor);

            if (prediction) {
              console.log(prediction);
            }
          } catch (e) {
            console.log(e);
          }
        }

        requestAnimationFrame(loop);
      };

      loop();
    };

Neither predict(), execute(), executeAsync() function on the model succeed to deliver any Result, instead I get this error

[Error: The shape of dict['ToFloat'] provided in model.execute(dict) must be [-1,-1,-1,3], but was [200,150,3]]

Like I said, I'm not really into TensorFlow but a Shape with negative Vector sizes does not seem quite right.

Would appreciate if someone has informations on whats wrong or missing.

Upvotes: 0

Views: 690

Answers (1)

Lescurel
Lescurel

Reputation: 11631

-1 for a dimension indicates that this dimension can vary. In your case ([-1,-1,-1,3]), the only constraints are the following:

  • Your input must have 4 dimensions
  • Your input's 4th dimension must be 3.

Your input only has 3 dimensions. You need to add a dimension to your input (presumably the batch dimension). You can use the expandDims function for that.

const prediction = await detectionModel.predict(nextImageTensor.expandDims(0));

Upvotes: 2

Related Questions