Yiming992 / LaneDetect

Deep Learning Lane Detection

Deep Learning Lane Detection

Layout：

|
|----README.md
|
|----model.py: Network definition
|
|----loss.py: implementation of discriminative loss
|
|----preprocess.py: create required dataset from uncompressed tusimple data
|
|----train.py: model training pipline
|
|----Data.py： pytorch dataset definition
|
|----clustering.py: posprocess procedures
|
|----logs/
|------|------models/: save serialized model pickle files
|------|------loggings/: save trainig logs
|
|----data/
|------|------cluster/: store instance segemntation ground-truth
|------|------train_binary/:store semantic-segmentation ground-truth
|------|------LaneImages/:store raw RGB images
|
|----test_result/
|------|-----fitted/:store test results with curve fitting 
|------|-----instance/:store instance segmentation test results
|
|----test_images/---store sample test images

Network architecture

Examples

Usage

predict on test images

1. Download trained pytorch model pickled from model
2. run test.py to produce segmented images in test_result/fitted and test_result/instance folders

train on tusimple data

1. Download tusimple dataset from link and decompress
2. run preprocess.py to create usable datasets
3. run train.py to perform training