This is the 7th place solution for the MosquitoAlert Challenge 2023. The goal of this competition is to identify mosquitoes and determine their species.
Install Datasets
data_round_2
(the annotations files are included).gbif-cropped
and inaturalist-six-cropped
(the annotations files are included).Install Dependencies
pip install -r requirements.txt
.Run the Classifier
experiments
directory and execute the following command: python mosquito_clf_yolo_lux_ema.py
.Install Competition Dataset
data_round_2
.Install Dependencies
pip install -r requirements.txt
.Prepare YOLO Dataset
experiments/yolo
directory and run the script: python convert_mosquito_to_yolo.py
.Start Training
python yolo_training.py
.You can find the model weights and instructions on how to use them on the Hugging Face Model Hub.
phase2_train_v0_cleaned.csv
was created using owl-vit
. You can refer to experiments/cleaning_annotations.ipynb
for details.phase2_train_v0_cleaned_yolo_best_annotations.csv
uses phase2_train_v0_cleaned.csv
along with YOLOv8-s model annotations. Refer to extra_data/annotate_images_yolo.py
for more information.best_model_val_data_yolo_annotations.csv
and best_model_train_data_yolo_annotations.csv
are train/validation splits of phase2_train_v0_cleaned_yolo_best_annotations.csv
.inaturalist.csv
contains annotations for lux's dataset. Since the images are already cropped, we used the entire image as the bounding box.