Spectral_Transformer

[Learning to Segment Complex Vessel-like Structures with Spectral Transformer,

usage

datasets

Download the datasets and the files follows the following structure.

new_dataset
|_test
|   |_images
|   |	|_CFD
|   |   |  |_CFD_00..jpg
|   |   |_cracktree
|   |   |  |_a..jpg
|   |_masks
|   |	|_CFD
|   |   |  |_CFD_00..jpg
|   |   |_cracktree
|   |   |  |_a..jpg
|_train
|   |_images
|   |	|_CFD
|   |   |  |_CFD_00..jpg
|   |   |_cracktree
|   |   |  |_a..jpg
|   |_masks
|   |	|_CFD
|   |   |  |_CFD_00..jpg
|   |   |_cracktree
|   |   |  |_a..jpg
|   |_CFD.txt
|   |_cracktree.txt


        ......

train.txt format

    train_img_dir = "./datasets/new_dataset/train/"+ datasetName +".txt"   
    valid_img_dir = "./datasets/new_dataset/test/images/"+datasetName+'/'
    valid_lab_dir = "./datasets/new_dataset/test/masks/"+datasetName+'/'
.....

Valid

Change the 'pretrain_dir','datasetName' and 'netName' in test.py

python test.py

Pre-trained model

Download the trained model.

|-- model
    |-- <netname>
        |   |-- <trained_model.pkl>
        ......

dataset	Pre-trained Model
CHASE-DB	[Link]
PRIME-FP20	[Link]
RECOVERY	[Link]
cracktree	[Link]
crackls315	[Link]
CFD	[Link]
Massachusetts Road	[Link]
DeepGlobe Road	[Link]

Acknowledgments

• We thank the anonymous reviewers for valuable and inspiring comments and suggestions.