Put the VOC2012 dataset in the following path:
data
|---VOCdevkit
|---VOC2012
|-------Annotations
|-------Imagesets
|-------JPEGImages
|-------SegmentationClass
|-------SegmentationObject
Put the Semantic Boundaries Dataset in the following path:
data
|---SBD
|---cls
|---img
|---inst
|---train.txt
|---val.txt
- Train on Semantic Boundaries Dataset (DeeplabV3+ for example)
python train.py --cfg experiments/deeplabv3plus_sbd.yaml
During training, you can use the command
tensorboard --logdir=runs
in the console to enter the tensorboard panel to visualize the training process.
- Resume training
python train.py --cfg experiments/deeplabv3plus_sbd.yaml --ckpt checkpoint_filepath
- Evaluate on Semantic Boundaries Dataset (DeeplabV3+ for example)
python train.py --cfg experiments/deeplabv3plus_sbd.yaml --mode valid --ckpt outputs/DeeplabV3Plus_SBD_weights.pth
- Test on single image or several images (DeeplabV3+ for example)
python test.py --cfg experiments/deeplabv3plus_sbd.yaml --ckpt outputs/DeeplabV3Plus_SBD_weights.pth