ml-sharp-onnx

ML-SHARP ONNX

ONNX conversion of Apple's SHARP single-image 3D Gaussian prediction model, for browser inference via ONNX Runtime Web + WebGPU.

Model Derivative Notice

This is a Model Derivative of Apple's SHARP model. The original PyTorch checkpoint has been converted to ONNX format using torch.onnx.export(). No retraining or fine-tuning was performed — the weights are numerically equivalent to the original checkpoint (sharp_2572gikvuh.pt).

Changes from the original model:

• Converted from PyTorch .pt format to ONNX (opset 18)
• Weights stored as external data (sharp.onnx.data) in fp32
• Model wrapped to return flat tensors instead of NamedTuple (required for ONNX compatibility)
• Fixed input shape: [1, 3, 1536, 1536] image + [1] disparity factor

Files

• sharp.onnx — ONNX computation graph (~7.4 MB)
• sharp.onnx.data — Model weights (~2.4 GB, fp32)

License

Apple Machine Learning Research Model is licensed under the Apple Machine Learning Research Model License Agreement.

This model is provided exclusively for Research Purposes — non-commercial scientific research and academic development activities. See the full license for details.

Attribution

Original model by Apple Inc. — apple/ml-sharp