DSACA

Code release for Dilated-Scale-Aware Category-Attention ConvNet for Multi-Class Object Counting (Accepted)

Changelog

• 2021/04/21 upload the code.

Requirements

• python 3.6
• PyTorch 1.3.0
• torchvision

Pre trained model

VisDrone_class8.pth and RSOC_class2.pth download.

• VisDrone model best (password:qsw6) Link.
• RSOC model best (will be released when the author is free).

Data

• Download datasets
• Extract them to dataset/VisDrone/ and dataset/RSOC/, respectively.

• e.g., VisDrone and RSOC (modified from the DOTA dataset) datasets

  -/DSACA-main
      -/DSACA-main/dataset
         └─── VisDrone
            └───VisDrone2019-DET-train
            └───VisDrone2019-DET-val
         └─── RSOC
            └───train
            └───val
            └───test_large-vehicle.txt
            └───test_ship.txt
            └───test_small-vehicle.txt
            └───train_large-vehicle.txt
            └───train_ship.txt
            └───train_small-vehicle.txt
      -/DSACA-main/pre_trained
         └─── VisDrone_class8.pth
         └─── RSOC_class2.pth
         └─── pre_trained.md
      -/DSACA-main/density_generate
         └─── RSOC_choose.py
         └─── VisDrone.py
         └─── RSOC.py
      -/DSACA-main/make_npydata
         └─── VisDrone_make_npydata.py
         └─── RSOC_make_npydata.py
      -/DSACA-main/Network
         └─── VisDrone_class8.py
         └─── baseline_DSAM_CAM.py
      -/DSACA-main/images
         └─── intro.png
      └─── config.py
      └─── dataset.py
      └─── image.py
      └─── utils.py
      └─── VisDrone_train_class8_CAM_DSAM.py
      └─── RSOC_train_class2_CAM_DSAM.py
      └─── README.md

Train & Test

• Cd density_generate then run RSOC_choose.py (choose large-vehicle and small-vehicle to vehicle) for multi-class scenario.
• Run VisDrone.py and RSOC.py for dataset pre-processing.
• Cd make_npydata then run VisDrone_make_npydata.py and RSOC_make_npydata.py for target path pre-saving.
• Edit config.py for training-parameters setting.
• Run VisDrone_train_class8_CAM_DSAM.py or RSOC_train_class2_CAM_DSAM.py for training & testing.

Contact

Thanks for your attention! If you have any suggestion or question, you can leave a message here or contact us directly:

• [email protected]
• [email protected]