Skip to content

Files

BiSeNetv2

BiSeNetv2

Realtime Semantic Segmentation on KITTI Dataset

Trained on CityScapes Dataset & Finetuned on KITTI Semantic dataset

Run Demo

# weight_path is default to checkpoints/BiseNetv2_150.pth
python3 demo.py --path PATH_TO_IMAGE --weight_path PATH_TO_BISENET_WEIGHTS

Screenshot 2021-06-01 00:36:28

Visualize KITTI dataset

python3 visualization.py

Finetuning

python3 train.py --datapath PATH_TO_DATASET --pretrained PATH_TO_WEIGHTS_TO_CONTINUE --batch_size 16

Testing

python3 test.py --weight_path PATH_TO_WEIGHTS

# you must have KITTI dataset in data/KITTI in the format below

Structure

    ├── checkpoints
		├── BiseNetv2_150.pth 	# path to model
	    ├── tensorboard 		# path to save tensorboard events
    ├── data 					# path to kitti semantic dataset
        ├── KITTI
          ├── testing
              ├── image_2
          ├── training
            ├── image_2
            ├── instance
            ├── semantic
            ├── semantic_rgb