Enhancing VoxelNext Model Performance

[haroon4412]

Hello,

I am currently using the VoxelNext model with nuscenes weights and I'm seeking ways to improve its performance. I have a few questions and would appreciate any guidance or suggestions:

1. Conversion to ONNX or TorchScript: Can the VoxelNext model be converted to ONNX or TorchScript formats? If so, could you give me some guidance or reference documentation on how to do this?
2. Performance Optimization: Are there other recommended strategies or practices for speeding up the VoxelNext model besides conversion to other formats? Any specific hardware or software configurations known to boost performance would be especially helpful.

[Petros626]

1. When the documentation did not contain this conversion, you have to search on your own.
2. search thorugh the PR, speeding up during inference or only conversion?