By Y. Di, S. L. Phung, J. Berg, J. Clissold and A. Bouzerdoum.
This repository is an official PyTorch implementation of TP-YOLO, published by ICIP 2023.
pip install .
yolo task=detect mode=predict model=weights/tp-yolo_kb.pt source=input/images save=True
Hyperparameters:
- Epochs: 450
- Batch size: 4
- Pre-train: None
Dataset:
Dataset | Paper | Class # | Image # | Instance # |
---|---|---|---|---|
Khapra beetle | ICIP2023 | 3 | 1,600 | 4,885 |
Pest24 | CEA2020 | 24 | 25,378 | 192,422 |
Command:
yolo task=detect mode=train model=cfg/tp-yolo.yaml data=datasets/pestdata.yaml epochs=450 batch=4
yolo task=detect mode=val model=weights/tp-yolo_kb.pt data=datasets/pestdata_val.yaml
Class | P | R | AP50 | AP |
---|---|---|---|---|
Khapra beetle (larvae) | 98.2 | 90.4 | 98.2 | 73.4 |
Khapra beetle (adult) | 99.4 | 99.5 | 99.5 | 85.0 |
Khapra beetle (skin) | 99.0 | 97.2 | 99.2 | 75.7 |
All | 98.8 | 95.7 | 98.9 | 78.0 |
The project is build upon ultralytics framework. We apprecate the contribution.
@inproceedings{tpyolo
title={TP-YOLO: A Lightweight Attention-based Architecture for Tiny Pest Detection},
author={Yang Di and Son Lam Phung and Julian van den Berg and Jason Clissold and Abdesselam Bouzerdoum},
booktitle={IEEE International Conference on Image Processing (ICIP)},
pages={3394-3398},
year={2023}
}