Please publish the "optimized ONNX" model referenced in the paper (Section 7.1) and/or pytorch

[esteves023]

In the PDF included in the repository, Section 7.1 Real-World Deployment states:

“Our experiments show that DeepInfant V2 can operate in near-real-time on mobile devices when deployed via optimized ONNX or CoreML formats.”

The repo already ships the Core ML (*.mlmodel) artifacts, but an ONNX version (or the PyTorch weights needed to export one) is not present. This prevents:

Running the model on Android or in a browser (TF.js / ONNX.js).

Using the included predict.py, which expects an ONNX model.

Reproducing the “near-real-time” results on non-Apple hardware.

Could you please:

1. Publish the optimized ONNX model for DeepInfant V2 (and, if possible, VGGish & AFP).

2. Alternatively, provide the PyTorch checkpoints (*.pth) so the community can export to ONNX with torch.onnx.export.

3. (Optional) Add a minimal export script or update the README to clarify the deployment workflow.

This would let researchers and parents without macOS/iOS evaluate DeepInfant exactly as described in the paper.