Skip to content

Latest commit

 

History

History
131 lines (106 loc) · 4.87 KB

DATA.md

File metadata and controls

131 lines (106 loc) · 4.87 KB
Raw

Dataset preparation

We provide the prepared datasets that you can directly download from this link

[Optional] If you want to prepare the dataset by yourself or test our codes on your dataset, you will need to set up the dataset with the following instructions.

Preparation

  • Download the pretrained scale prediction model (pretrained_scale_prediction_model.pth) from the link.
  • Place the model on the root: $IIM/datasets/dataset_prepare

Shanghai Tech Part A

  • Download the images (train_data, test_data) from [Link].

  • Place the train_data and test_data to make the data folder like:

    --ProcessedData
    |-- SHHA
     -- |-- train_data
        |-- test_data
        |-- images               # To be generated
        |-- masks                # To be generated
        |-- size_map             # To be generated
        |-- jsons                # To be generated
        |-- train.txt            # To be generated
        |-- val.txt              # To be generated
        |-- test.txt             # To be generated
        |-- val_gt_loc.txt       # To be generated
        |-- test_gt_loc.txt      # To be generated

  • To generate other folders and files, you should run the command:

    cd $IIM/datasets/dataset_prepare
python prepare_SHHA.py

    To generate the required files for training and testing, you should make sure the following steps are completed successfully.

    if __name__ == '__main__':
    #================1. resize images ===================
    resize_images(train_path, 0)
    resize_images(test_path, 300)

    # ================2. size_map ==================
    from datasets.dataset_prepare.scale_map import main
    main ('SHHA')

    # ================3. box_level annotations ==================
    writer_jsons()

    # ================4. masks ==================
    generate_masks()

    # ================5. train test val id==================
    divide_dataset()
    
    # ==============6. generate val_loc_gt.txt and test_loc_gt.txt==================
    loc_gt_make(mode = 'test')
    loc_gt_make(mode='val')

    print("task is finished")
    ~~~

Shanghai Tech Part B

  • Download the images (train_data, test_data) from [Link].

  • Place the train_data and test_data to make the data folder like the Shanghai Tech Part A dataset.

  • Run

    cd $IIM/datasets/dataset_prepare
python prepare_SHHB.py

UCF-QNRF

  • Download the images (Train, Test) from [Homepage] or [Download].

  • Place the Train and Test to make the data folder like the Shanghai Tech Part A dataset:

  • Run

    cd $IIM/datasets/dataset_prepare
python prepare_QNRF.py

JHU

  • Download the images (train_data, test_data) from [Homepage].

  • Place the train_data and test_data to make the data folder like.

    --ProcessedData
    |-- JHU
     -- |-- train
        |-- val
        |-- test
        |-- images               # To be generated
        |-- masks                # To be generated
        |-- jsons                # To be generated
        |-- train.txt            # To be generated
        |-- val.txt              # To be generated
        |-- test.txt             # To be generated
        |-- val_gt_loc.txt       # To be generated
        |-- test_gt_loc.txt      # To be generated

  • To generate other folders and files, you should run the command:

    cd $IIM/datasets/dataset_prepare
python prepare_JHU.py

    To generate the required files for training and testing, make sure the following steps are completed successfully.

    if __name__ == '__main__':
#================1. resize images and gt===================
    resize_images('train')
    resize_images('val')
    resize_images('test')

    # ================2. masks==================
    generate_masks()

    # ================3. train test val id==================
    JHU_list_make('test')
    JHU_list_make('val')
    JHU_list_make('train')

    # ================4. generate val_loc_gt.txt and test_loc_gt.txt==================
    loc_gt_make(mode = 'test')
    loc_gt_make(mode='val')

    print("task is finished")
    ~~~