search

vikolss / DACS

Code from the paper "DACS: Domain Adaptation via Cross-domain Mixed Sampling"
MIT License
123 stars 19 forks source link

Pretrained weight or Code for SYNTHIA to Cityscapes transfer #4

Closed bobcheng15 closed 3 years ago

bobcheng15 commented 4 years ago

Hi,

Thank you for your amazing work. I have successfully ran gta->cityscapes version of your repository, but have difficulties running it on SYNTHIA->Cityscapes. Can you please instruct me on how to run the SYNTHIA to Cityscapes transformation or release the weight for this transform ?

Thank you very much!

WilhelmT commented 4 years ago

To train Synthia you simply need to change the following lines in trainUDA.py.

From:

    #New loader for Domain transfer
    if True:
        data_loader = get_loader('gta')
        data_path = get_data_path('gta')
        if random_crop:
            data_aug = Compose([RandomCrop_gta(input_size)])
        else:
            data_aug = None

        #data_aug = Compose([RandomHorizontallyFlip()])
        train_dataset = data_loader(data_path, list_path = './data/gta5_list/train.txt', augmentations=data_aug, img_size=(1280,720), mean=IMG_MEAN)

to

    #New loader for Domain transfer
    if True:
        data_loader = get_loader('synthia')
        data_path = get_data_path('synthia')
        if random_crop:
            data_aug = Compose([RandomCrop_gta(input_size)])
        else:
            data_aug = None

        #data_aug = Compose([RandomHorizontallyFlip()])
        train_dataset = data_loader(data_path, list_path = './data/synthia_list/train.txt', augmentations=data_aug, img_size=(1280,720), mean=IMG_MEAN)