Converting RTDETR Object Detection Model to CoreML

[ProfessorHT]

I am trying to convert the RTDETR Object Detection model to CoreML for use in an iOS app. However, I am encountering a data type mismatch error during the conversion process.

The code i am using:

import torch 
import coremltools as ct
from transformers import RTDetrForObjectDetection

model = RTDetrForObjectDetection.from_pretrained("TMP-finetune/runs/Jan22_16-22-29")

example_input = (torch.rand(1, 3, 512, 512),)

# Generate ExportedProgram by exporting
exported_model = torch.export.export(model, example_input)

input_images = ct.ImageType(name="images", shape=example_input[0].shape)

# Convert the model
mlmodel = ct.convert(
    exported_model,
    inputs=[input_images]
)

# Save the Core ML model
mlmodel.save("best.mlmodel")

The error i am getting:

I have tried searching for solutions but haven't found a way to resolve this data type mismatch issue. Any guidance or suggestions would be greatly appreciated.

I have also opened a related issue on another repository, which can be found here: huggingface/transformers#35905 (comment)

Thanks in advance