I succesfully converted and loaded efficientdet model feature vector from TensorFlow Hub finetuned using approach from automl/efficientdet at master · google/automl · GitHub.
I was able to run model on python (correctly detects objects) and to convert it using tensorflow==2.12.0 and coremltools==6.3.0 (to both: “mlprogram” and “neuralnetwork” with same result) to coreML supported model format.
Within the coreML model is loaded with no errors, it’s metadata is correctly parsed, but during prediction following error occurs:
2023-06-05 09:10:56.942952+0200 xxxx[1619:435597] [espresso] [Espresso::handle_ex_plan] exception=Espresso exception: "Invalid state": Stack_nd layer: Invalid shapes of input tensors. status=-5
2023-06-05 09:10:56.943016+0200 xxxx[1619:435597] [coreml] Error computing NN outputs -5
2023-06-05 09:10:56.943088+0200 xxxx[1619:435597] [coreml] Failure in -executePlan:error:.
Unable to classify image.
The VNCoreMLTransform request failed
Error: Vision request failed with error "Error Domain=com.apple.vis Code=3 "The VNCoreMLTransform request failed" UserInfo={NSLocalizedDescription=The VNCoreMLTransform request failed, NSUnderlyingError=0x2801267f0 {Error Domain=com.apple.CoreML Code=0 "Error computing NN outputs." UserInfo={NSLocalizedDescription=Error computing NN outputs.}}}"
2023-06-05 09:10:56.991190+0200 xxxx[1619:435607] [espresso] [Espresso::handle_ex_plan] exception=Espresso exception: "Invalid state": Stack_nd layer: Invalid shapes of input tensors. status=-5
2023-06-05 09:10:56.991256+0200 xxxx[1619:435607] [coreml] Error computing NN outputs -5
2023-06-05 09:10:56.991285+0200 xxxx[1619:435607] [coreml] Failure in -executePlan:error:.
Unable to classify image.
The VNCoreMLTransform request failed
Has anyone managed to run efficientdet-lite0 or any other model from this repository on iOS?
Any help or suggestion would be really appreciated.
Threads created in coreML forum: Error during prediction with conve… | Apple Developer Forums
Issue created in google/autoML forum: Error during prediction within coreML framework of the converted Efficientdet-lite0 model · Issue #1195 · google/automl · GitHub