| |
| """CPU reference of BEVFusion-L (Autoware autoware_bevfusion, lidar-only) -- the ground truth every PCC / output- |
| agreement gate compares against (PLAN.md 0.3 item 1). It runs where ttnn is absent (the research venv) and never |
| imports ttnn. |
| |
| - ``config.py`` the Autoware parameters this port reproduces (the ml_package / remapper / metadata yaml of the |
| weights, ``bevfusion_lidar.param.yaml``, the node's sanitization) and the port's COMPILE constants |
| (256,000 voxels, the sparse level grids, the per-level capacity buckets, top-k 512), each with its |
| source line in autoware_universe @ 9ceaccf026c31ffc5319bc9eeb4bd7bede0af3fd; |
| - ``weights.py`` ``bevfusion_lidar.onnx`` read as data with the vendored ``ttaw.weights.OnnxWeights`` |
| (initializers and constants by consuming node; every wiring assumption asserted from the graph, |
| including the 21 sparse convs' ``GetIndicePairsImplicitGemm`` attributes; all 44 BatchNorms |
| folded in float64 where the TT port folds them); |
| - ``model.py`` fp32 torch rebuild of the deployed graph (gather-form sparse encoder on the host rulebooks, |
| to-dense, SECOND, SECONDFPN, dense head, local max on classes 0-3, ONNX-semantics top-K with |
| optional teacher forcing, query init, decoder, prediction heads, output assembly) with per-module |
| taps (``ttaw.golden.TapRegistry``; names in ``MODULE_TAPS``); |
| - ``ort.py`` ONNX Runtime on the plugin-free dense sub-graph of the shipped ONNX, with optional intermediates |
| (research / tests only: onnxruntime is not in the image; the full graph needs TensorRT plugins); |
| - ``pipeline.py`` the reference end to end: ``host/`` pre-processing -> torch network (or torch sparse + ORT |
| dense) -> ``host/`` post-processing -> ``ttaw.outputs.Detections3D``. |
| |
| Pre- and post-processing are not duplicated here: they live in ``tt_bevfusion.host`` (shared with the device path, so |
| TT and CPU see identical voxel, feature and rulebook tensors) and are tested against the research scripts. |
| |
| The ``to_dict()`` of this reference's output on the shipped sample is stored as |
| ``samples/test_pcd.reference.json``: ``server/smoke_test.py`` compares the served output with it. |
| """ |
|
|
