How nesl/radar-camera-fusion-depth works
Project type and purpose
• This is a research‐oriented, two‐stage deep‐learning pipeline for depth estimation, fusing camera images and mmWave radar point clouds.
• The code handles dataset preparation (nuScenes), two model stages (RadarNet and FusionNet), training, evaluation, and visualization.
File structure
• Top‐level directories:
• bash/ – shell scripts for data setup, training, and evaluation
• setup/ – Python scripts to preprocess raw nuScenes data into training/test sets
• src/ – core code: dataset classes, transforms, model definitions (radarnet_model.py, fusionnet_model.py), training and inference entrypoints
• tools/ – visualization utilities
• Key patterns: a clear “data preparation → stage1 model → stage2 model → evaluation/visualization” pipeline
• Configuration: requirements.txt, license, and README guide installation and high‐level flow
README insights
• Describes sequence: environment setup → dataset prep → train RadarNet → dataset prep for FusionNet → train FusionNet → evaluate
• Emphasizes use of nuScenes, PyTorch, TensorBoard, and provided shell wrappers