hengweibin/centerfusiondetect3d

How hengweibin/centerfusiondetect3d works

CenterFusionDetect3D implements camera–radar fusion for nuScenes 3D object detection. The documented workflow converts nuScenes annotations, loads image and radar samples, trains or evaluates a CenterFusion model, then decodes and presents detections; separate demo and vision-inference entry points are also documented. The graph includes documented evaluation and export capabilities, but their runtime wiring is not fully shown. Edges follow the excerpts, SOURCE INDEX references, or README; unsampled details are omitted.

Main entry points: main.py Train and evaluate · detector.py Detector · evaluate.py nuScenes evaluation · nuscenes.py nuScenes dataset

Components 27

Data and workflows

Configuration src/lib/config/default.py
Dataset conversion src/convert_nuScenes.py
nuScenes dataset src/lib/dataset/datasets/nuscenes.py
Sample preparation src/lib/dataset/generic_dataset.py
Image transforms src/lib/utils/image.py
Radar geometry src/lib/utils/pointcloud.py

Detection model

DLA backbone src/lib/model/networks/dla.py
Radar fusion src/lib/model/networks/fusionModules.py
Detection heads src/lib/model/networks/detectHeads.py
CenterFusion model src/lib/model/model.py
Detection losses src/lib/model/losses.py
Loss aggregation src/lib/model/genericLoss.py
Output decoding src/lib/model/decode.py

Training and inference

Trainer src/lib/trainer.py
Train and evaluate src/main.py
Detector src/lib/detector.py
3D postprocessing src/lib/utils/postProcess.py
Visualization demo src/demo.py
Vision inference src/inference.py

nuScenes tools

nuScenes evaluation src/lib/nuScenes_lib/evaluate.py
Evaluation loaders src/lib/nuScenes_lib/loaders.py
Evaluation algorithms src/lib/nuScenes_lib/algo.py
KITTI export src/lib/nuScenes_lib/export_kitti.py
Radar utilities src/lib/nuScenes_lib/utils_radar.py
KITTI utilities src/lib/nuScenes_lib/utils_kitti.py

Other components

Researcher
nuScenes data

Connections 28

Researcher → Train and evaluate: trains or evaluates
Researcher → Visualization demo: runs demo
Researcher → Vision inference: runs inference
nuScenes data → Dataset conversion: converts dataset
Dataset conversion → nuScenes dataset: prepares annotations
Train and evaluate → Configuration: loads settings src/main.py
Train and evaluate → nuScenes dataset: gets dataset src/main.py
Train and evaluate → CenterFusion model: creates model src/main.py
Train and evaluate → Trainer: starts training src/main.py
nuScenes dataset → Sample preparation: inherits from src/lib/dataset/datasets/nuscenes.py
Sample preparation → Image transforms: transforms images src/lib/dataset/generic_dataset.py
Sample preparation → Radar geometry: processes radar src/lib/dataset/generic_dataset.py
CenterFusion model → DLA backbone: builds backbone src/lib/model/model.py
DLA backbone → Radar fusion: uses fusion modules
DLA backbone → Detection heads: uses detection heads
Trainer → Loss aggregation: uses loss wrapper
Loss aggregation → Detection losses: computes losses src/lib/model/genericLoss.py
Visualization demo → Detector: creates detector
Vision inference → Detector: uses detector src/inference.py
Detector → CenterFusion model: runs model src/lib/detector.py
Detector → Output decoding: decodes outputs
Detector → 3D postprocessing: postprocesses detections src/lib/detector.py
nuScenes evaluation → Evaluation loaders: uses loaders src/lib/nuScenes_lib/evaluate.py
nuScenes evaluation → Evaluation algorithms: uses algorithms src/lib/nuScenes_lib/evaluate.py
nuScenes evaluation → nuScenes dataset: uses dataset definitions src/lib/nuScenes_lib/evaluate.py
KITTI export → KITTI utilities: uses utilities src/lib/nuScenes_lib/export_kitti.py
Researcher → nuScenes evaluation: evaluates results
nuScenes data → Evaluation loaders: supplies evaluation data

More Python diagrams: pollen-robotics/reachy_mini · massive-com/client-python · bramstroker/homeassistant-powercalc · wntrblm/nox · xilinx/brevitas

Drawn by GitDiagram on September 30, 2026 from the repository's file tree, README and source files. Read as Markdown ·