Drivable Area Segmentation

The repository is used to train a drivable-area-segmentation model with deeplabv3+, and the trained model can inference with libtorch. The repository is based on deeplabs and Person_Segmentation.

Data Preparation

Please prepare dataset under the data directory. You can download from link：https://pan.baidu.com/s/1NT5-Ldo0kYIfCcUbkq0xEA password：76bi

Generate labels for custom dataset. The test samples are under the samples directory, and generated labels are under the results directory.
```
python hand_seg.py 
```

Pretrained Model

Please download pretrained model under the weights directory. Link：https://pan.baidu.com/s/1l78GrdJ8FoghqfzGGw0Kng password：vm33

Train

python train.py

Inference with python

You download the trained model from link(https://pan.baidu.com/s/1rnOtdmewT-oho-cU6bLluw password：djue) and place it under the weights directory. The test samples are under the samples directory, and inferece results are under the results directory.

python inference.py

Inference with c++

python generate_half_model.py
cd  LibtorchSegmentation/build
cmake ..
make
./drive