Skip to content
/ TADA Public

πŸš› TrAsh DAtaset based on Korea Ministry of Environment guidelines

License

Notifications You must be signed in to change notification settings

jms0923/TADA

Folders and files

NameName
Last commit message
Last commit date

Latest commit

Β 

History

4 Commits
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 

Repository files navigation

TADA is large-scale high-resolution trash datasets for solving trash-detection problems. It was intended for general household goods. And dataset was classified based on guidelines for separating and discharging recyclables from the Ministry of Environment. There were 10 categories, and 33,434 objects were annotated for 4977 images with 1280 x 720 resolution.

For convenience, annotations are provided in COCO data format. If you want to know about more detail Data format reference COCO data format

Models

AP performance by Category of each model

Model mAP Config Checkpoint
SSD 0.310 Download Download
YOLO v3 0.445 Download Download
Cascade RCNN 0.432 Download Download

Getting started

  1. Install mmdetection

    This implementation is based on mmdetection. Please refer to INSTALL.md for installation and dataset preparation.

  2. Copy the codes to mmdetection directory

    cp -r configs/ ${MMDETECTION_PATH}/
  3. Prepare data

    Download from DOI

    The directories should be arranged like this:

    >   mmdetection
    >     β”œβ”€β”€ mmdet
    >     β”œβ”€β”€ tools
    >     β”œβ”€β”€ configs
    >     β”œβ”€β”€ data
    >     β”‚   β”œβ”€β”€ coco
    >     β”‚   β”‚   β”œβ”€β”€ annotations
    >     β”‚   β”‚   β”œβ”€β”€ train2017
    >     β”‚   β”‚   β”œβ”€β”€ val2017
    >     β”‚   β”‚   β”œβ”€β”€ test2017
    
  4. Train with GPUs

    CONFIG_FILE=configs/ ${CONFIG_FILE}
    ./ tools/dist_train.py ${CONFIG_FILE} ${NUM_GPUs}
  5. Calculate parameters and flops

     python tools/get_flops.py ${CONFIG_FILE} --shape $SIZE $SIZE
  6. Test

    python tools/test.py ${CONFIG_FILE} ${CHECKPOINT_FILE} --out  ${OUTPUT_FILE} --eval bbox

More usages can reference mmdetection documentation.

TODO

  • Release pre-train models.
  • Update our arxiv paper.

Citation

If you find our research useful, please cite the paper:

@inproceedings{sun2018fishnet,
  title={Large-Scale High-Resolution Dataset for Trash Detection},
  author={Min-Seok Jo, Seong-Soo Han, Chang-Sung Jeong},
  booktitle={},
  pages={--},
  year={2021}
}

Contact

You can contact MinSeok Jo by sending email to [email protected]


inference results