Object detection dataset:
https://www.kaggle.com/datasets/eashenyang/mosquito-alert-2023-detection
Image classification dataset:
https://www.kaggle.com/datasets/eashenyang/mosquito-alert-2023-classification
Object detection dataset: train/detection/dataset
dataset
├── aegypti
├── albopictus
└── yolo_f0_raw
Image classification dataset: train/classification/dataset
dataset
├── origin_dataset
└── inaturalist
cd train/detection
sh train.sh
get your best model under train/detection/runs/train
directory
Download pretrain model from following link
https://drive.google.com/file/d/17sUNST7ivQhonBAfZEiTOLAgtaHa4F3e/view?usp=sharing
Place metafg_2_inat21_384.pth
at train/classification
Using following commands, you will get 10 models from 10 train/val splits
cd train/classification
sh batch_train.sh
Using following commands, you will get uniform soup model from 10 models
cd train/classification
python avg_ckpt.py {your_args}
The original dataset underwent the removal of images containing multiple mosquitoes, and bounding boxes were manually refined. Additionally, an additional dataset was compiled from iNaturalist, featuring Aedes aegypti and Aedes albopictus species, with mosquito bounding boxes also being manually labeled.
In the original dataset, images with noisy labels were eliminated. An additional dataset was acquired from iNaturalist, and the mosquito class for each entry was meticulously verified through manual inspection.