chips infrastructure
The Compressed Model Never Needed Its Source Code
H3DNAS searches and rewrites ONNX graphs directly for edge deployment of 3D point-cloud networks.
Summary
H3DNAS searches and rewrites ONNX graphs directly for edge deployment of 3D point-cloud networks.
A channel-dependency graph classifies ONNX operators and computes a topology-defined ceiling on how much of a model can be pruned. A two-stage search then selects channels by importance, ranks candidates by output fidelity without labels and mutates Pareto candidates with GhostConv. On ModelNet40, the paper reports parameter reductions of 43.2% to 65.5% and Jetson Orin Nano speedups of 1.29× to 1.99× across three architectures with negligible accuracy loss under its evaluation.
Why it matters
H3DNAS searches and rewrites ONNX graphs directly for edge deployment of 3D point-cloud networks.
Limits and context
No additional limitation was separately recorded.
Key claims
H3DNAS searches and rewrites ONNX graphs directly for edge deployment of 3D point-cloud networks.
Evidence: source-2026-09-03-007
Sources
- arXiv preprint 2609.02684arXiv · primary research
Corrections
No corrections have been recorded for this story.