Skip to content

decisionforce/VPN

Repository files navigation

Cross-view Semantic Segmentation for Sensing Surroundings

We release the code of the View Parsing Networks, the main model for Cross-view Semantic Segmentation task.

Cross-View Semantic Segmentation for Sensing Surroundings
Bowen Pan, Jiankai Sun, Ho Yin Tiga Leung, Alex Andonian, and Bolei Zhou
IEEE Robotics and Automation Letters
In IEEE International Conference on Intelligent Robots and Systems (IROS) 2020
[Paper] [Project Page]

@ARTICLE{pan2019crossview,
     author={B. {Pan} and J. {Sun} and H. Y. T. {Leung} and A. {Andonian} and B. {Zhou}},
     journal={IEEE Robotics and Automation Letters},
     title={Cross-View Semantic Segmentation for Sensing Surroundings},
     year={2020},
     volume={5},
     number={3},
     pages={4867-4873},
}

Requirement

  • Install the House3D simulator, or Gibson simulator.
  • Software: Ubuntu 16.04.3 LTS, CUDA>=8.0, Python>=3.5, PyTorch>=0.4.0

Train and test VPN

Data processing (use House3D for example)

Training Command

# Training in indoor-room scenarios, using RGB input modality, with 8 input views.
python -u train.py --fc-dim 256 --use-depth false --use-mask false --transform-type fc --input-resolution 400 --label-res 25 --store-name [STORE_NAME] --n-views 8 --batch-size 48 -j 10 --data_root [PATH_TO_DATASET_ROOT] --train-list [PATH_TO_TRAIN_LIST] --eval-list [PATH_TO_EVAL_LIST]

# Training in driving-traffic scenarios, using RGB input modality, with 6 input views.
python -u train_carla.py --fc-dim 256 --use-depth false --use-mask false --transform-type fc --input-resolution 400 --label-res 25 --store-name [STORE_NAME] --n-views 6 --batch-size 48 -j 10 --data_root [PATH_TO_DATASET_ROOT] --train-list [PATH_TO_TRAIN_LIST] --eval-list [PATH_TO_EVAL_LIST]

Testing Command

# Training in indoor-room scenarios, using RGB input modality, with 8 input views.
python -u test.py --fc-dim 256 --use-depth false --use-mask false --transform-type fc --input-resolution 400 --label-res 25 --store-name [STORE_NAME] --n-views 8 --batch-size 4 --test-views 8 --data_root [PATH_TO_DATASET_ROOT] --eval-list [PATH_TO_EVAL_LIST] --num-class [NUM_CLASS] -j 10 --weights [PATH_TO_PRETRAIN_MODEL]

# Testing in driving-traffic scenarios, using RGB input modality, with 6 input views.
python -u test_carla.py --fc-dim 256 --use-depth false --use-mask false --transform-type fc --input-resolution 400 --label-res 25 --store-name [STORE_NAME] --n-views 6 --batch-size 4 --test-views 6 --data_root [PATH_TO_DATASET_ROOT] --eval-list [PATH_TO_EVAL_LIST] --num-class [NUM_CLASS] -j 10 --weights [PATH_TO_PRETRAIN_MODEL]

Transfer learning for sim-to-real adaptation

Data processing (use indoor-room scenarios for example)

Training Command

# Training in indoor-room scenarios, using RGB input modality, with 8 input views.
python -u train_transfer.py --task-id [TASK_NAME] --num-class [NUM_CLASS] --learning-rate-D 3e-6 --iter-size-G 1 --iter-size-D 1 --snapshot-dir ./snapshot --batch-size 20 --tensorboard true --n-views 6 --train_source_list [PATH_TO_TRAIN_LIST] --train_target_list [PATH_TO_EVAL_LIST] --VPN-weights [PATH_TO_PRETRAINED_WEIGHT] --scenarios indoor

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages