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
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 |
-
Install mmdetection
This implementation is based on mmdetection. Please refer to INSTALL.md for installation and dataset preparation.
-
Copy the codes to mmdetection directory
cp -r configs/ ${MMDETECTION_PATH}/
-
Prepare data
The directories should be arranged like this:
> mmdetection > โโโ mmdet > โโโ tools > โโโ configs > โโโ data > โ โโโ coco > โ โ โโโ annotations > โ โ โโโ train2017 > โ โ โโโ val2017 > โ โ โโโ test2017
-
Train with GPUs
CONFIG_FILE=configs/ ${CONFIG_FILE} ./ tools/dist_train.py ${CONFIG_FILE} ${NUM_GPUs}
-
Calculate parameters and flops
python tools/get_flops.py ${CONFIG_FILE} --shape $SIZE $SIZE
-
Test
python tools/test.py ${CONFIG_FILE} ${CHECKPOINT_FILE} --out ${OUTPUT_FILE} --eval bbox
More usages can reference mmdetection documentation.
- Release pre-train models.
- Update our arxiv paper.
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}
}
You can contact MinSeok Jo by sending email to [email protected]