Skip to content

fundamentalvision/BEVFormer

Embedl's build embedl/bevformer-small of the original model fundamentalvision/BEVFormer

BEVFormer-small architecture graph
BEVFormer-small architecture graph
  • fundamentalvision/BEVFormer
  • FP16
  • L4
Labels
  • Image
  • Detection
Build
Baseline
Precision
FP16
Runtime
TensorRT 8.5.1.7 with BEVFormer_tensorrt plugins
Cameras
6 views, 1280×736 each
BEV grid
150×150
Run BEVFormer-small on L4

BEVFormer-small (R101-DCN backbone) detects 3D objects in the ten nuScenes classes from six surround-view cameras.

BEVFormer is by Zhiqi Li et al. (arXiv:2203.17270), Apache-2.0, at fundamentalvision/BEVFormer; the TensorRT export and plugins are from DerryHub/BEVFormer_tensorrt, Apache-2.0.

Methodology

  • Latency: trtexec GPU compute time of the TensorRT 8.5.1.7 FP16 engine on the NVIDIA L4, batch 1, one nuScenes sample (six cameras); mean, p99 and throughput from the same run.
  • Accuracy: NDS and mAP on the 6,019-sample nuScenes validation set, evaluated in PyTorch FP16 on an NVIDIA L4 with BEVFormer_tensorrt’s bevformer_small_trt_p config, not on the device engine. The BEVFormer model zoo’s numbers are shown beside it.
  • Build: exported to ONNX with DerryHub/BEVFormer_tensorrt and compiled on the device with trtexec --fp16 and its plugin library, which the runtime image carries. No Embedl optimization.

Files

Artifacts

What ships in the model bundle.

File Kind Size sha256
TensorRT engine

engine/bevformer_small_trt_p_cp_trtexec.engine

Engine 124.0 MiB 4b6241cf5346
trtexec build log

logs/trtexec_build.log

Log 5.9 MiB 38a38f2d04d5
HTTP server for the engine (serve.sh --port 8000)

engine/serve.sh

script 2 KiB c59eaebcab93
Model settings

bevformer.json

Config 394 B bd79c1d1af32
Latency (trtexec)

benchmark/metrics.json

Results 4 KiB 664b85c342c8
Accuracy (PyTorch, on the host)

benchmark/accuracy.json

Results 269 B e28f25d1f5bf
nuScenes sample (engine inputs, .npz)

inference/nuscenes_sample.npz

sample 86.7 MiB a261a50479a5
Detections for the sample

inference/detections.json

Results 101 KiB 9248aab2e23f
BEVFormer_tensorrt licence

LICENSE-BEVFormer_tensorrt

license 11 KiB c71d239df917

Variants

Every build of BEVFormer-small on L4.

Variant Precision Speed Runtime
Baseline This page
FP16 116.73 ms
TensorRT 8.5.1.7 with BEVFormer_tensorrt plugins
bevformer.l4.runtime.amd64.202610130739ec55
INT8 111.91 ms
TensorRT 8.5.1.7 with BEVFormer_tensorrt plugins and Embedl optimizations
bevformer.l4.runtime.amd64.202610130739ec55