anilbatra2185/road_connectivity

spacenet3
│
└───AOI_2_Vegas_Train
│   └───RGB-PanSharpen
│   └───geojson
│   └───summaryData
│
└───AOI_3_Paris_Train
│   └───RGB-PanSharpen
│   └───geojson
│   └───summaryData
|
└───AOI_4_Shanghai_Train
|   .
|
└───AOI_5_Khartoum_Train
|   .
|

cd preprocessing
bash prepare_spacenet.sh /spacenet3

spacenet3
|
└───full
│   └───gt
│   └───images

deepglobe
│
└───train
│   └───gt
│   └───images

bash split_data.sh /spacenet3/full /data/spacenet/ .png .png
bash split_data.sh /deepglobe/train /data/deepglobe _sat.jpg _mask.png

data/spacenet
|   train.txt
|   val.txt
|   train_crops.txt   # created by script
|   val_crops.txt     # created by script
|
└───train
│   └───gt
│   └───images
└───val
│   └───gt
│   └───images
└───train_crops       # created by script
│   └───gt
│   └───images
└───val_crops         # created by script
│   └───gt
│   └───images

python create_crops.py --base_dir /data/spacenet/ --crop_size 650 --crop_overlap 215 --im_suffix .png --gt_suffix .png
python create_crops.py --base_dir /data/deepglobe/ --crop_size 512 --crop_overlap 256 --im_suffix _sat.jpg --gt_suffix _mask.png

usage: train_mtl.py [-h] --config CONFIG
                    --model_name {LinkNet34MTL,StackHourglassNetMTL}
                    --dataset {deepglobe,spacenet}
                    --exp EXP
                    [--resume RESUME]
                    [--model_kwargs MODEL_KWARGS]
                    [--multi_scale_pred MULTI_SCALE_PRED]

optional arguments:
  -h, --help            show this help message and exit
  --config CONFIG       config file path
  --model_name 			{LinkNet34MTL,StackHourglassNetMTL}
                        Name of Model = ['StackHourglassNetMTL',
                        'LinkNet34MTL']
  --exp EXP             Experiment Name/Directory
  --resume RESUME       path to latest checkpoint (default: None)
  --dataset 			{deepglobe,spacenet}
                        select dataset name from ['deepglobe', 'spacenet'].
                        (default: Spacenet)
  --model_kwargs 		MODEL_KWARGS
                        parameters for the model
  --multi_scale_pred 	MULTI_SCALE_PRED
                        perform multi-scale prediction (default: True)

CUDA_VISIBLE_DEVICES=0,1 python train_mtl.py --config config.json --dataset deepglobe --model_name "StackHourglassNetMTL" --exp dg_stak_mtl

CUDA_VISIBLE_DEVICES=0,1 python train_mtl.py --config config.json --dataset deepglobe --model_name "LinkNet34MTL" --exp dg_L34_mtl --multi_scale_pred false

CUDA_VISIBLE_DEVICES=0,1 python train_refine_pre.py --config config.json --dataset spacenet --model_name "LinkNet34" --exp spacenet_L34_pre_train_with_corruption --multi_scale_pred false

@InProceedings{Batra_2019_CVPR,
	author = {Batra, Anil and Singh, Suriya and Pang, Guan and Basu, Saikat and Jawahar, C.V. and Paluri, Manohar},
	title = {Improved Road Connectivity by Joint Learning of Orientation and Segmentation},
	booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
	month = {June},
	year = {2019}
}