How wvangansbeke/lanedetection_end2end works
This repository implements two end-to-end lane-detection variants: fitting lane curves in bird’s-eye view and backprojecting them to score coordinates in the original perspective. The principal workflow loads TuSimple images and annotations, predicts lane weight maps, fits curves with differentiable weighted least squares, and trains against geometric losses; evaluation reports lane accuracy and false-positive/negative rates. The README also describes segmentation and proxy-task options. The graph preserves those capabilities at the documented module level; omitted source excerpts limit certainty about some optional wiring.
Main entry points: Birds_Eye_View_Loss/main.py BEV training · Backprojection_Loss/main.py Backprojection training · LSQ_layer.py BEV network · Least_squares_net.py Backprojection network
Components 17
Bird’s-eye fitting
- BEV training
Birds_Eye_View_Loss/main.py - BEV loader
Birds_Eye_View_Loss/Dataloader/Load_Data_new.py - BEV network
Birds_Eye_View_Loss/Networks/LSQ_layer.py - ERFNet backbone
Birds_Eye_View_Loss/Networks/ERFNet.py - BEV losses
Birds_Eye_View_Loss/Loss_crit.py - BEV evaluation
Birds_Eye_View_Loss/eval_lane.py
Backprojection fitting
- Backprojection training
Backprojection_Loss/main.py - Backprojection loader
Backprojection_Loss/Dataloader/Load_Data_new.py - Backprojection network
Backprojection_Loss/Networks/Least_squares_net.py - Backprojection fitting
Backprojection_Loss/Networks/LSQ_layer.py - Backprojection losses
Backprojection_Loss/Loss_crit.py - Backprojection evaluation
Backprojection_Loss/eval_lane.py
Data and evaluation
- Curve labels
Birds_Eye_View_Loss/Labels - Lane labels
Backprojection_Loss/Labels - Predictions and scores
Other components
- Researcher
- TuSimple data
Connections 20
- Researcher → BEV training: starts training
- Researcher → Backprojection training: starts training
- TuSimple data → BEV loader: provides images and labels
- TuSimple data → Backprojection loader: provides images and labels
- Curve labels → BEV loader: supplies annotations
- Lane labels → Backprojection loader: supplies annotations
- BEV training → BEV loader: creates loaders
- BEV training → BEV network: initializes and trains
- BEV network → ERFNet backbone: uses backbone
- BEV training → BEV losses: computes objectives
- BEV training → BEV evaluation: evaluates lanes
- BEV training → Predictions and scores: writes model and metrics
- Backprojection training → Backprojection loader: creates loaders
- Backprojection training → Backprojection network: initializes and trains
- Backprojection network → Backprojection fitting: uses fitting layer
- Backprojection training → Backprojection losses: computes objectives
- Backprojection training → Backprojection evaluation: evaluates lanes
- Backprojection training → Predictions and scores: writes model and metrics
- BEV network → BEV losses: produces fitted curves
- Backprojection network → Backprojection losses: produces fitted curves