OSEA Mobile

Flutter app for offline bird identification

For the detecting task, we uses pretrained model ssd mobilenet.

For the classification task, ResNet34 model structure was adopted, take MetaFGNet/L_Bird_pretrain/checkpoints as a pretrained model, trained on DongNiao DIB-10K using Gorilla-Lab-SCUT/MetaFGNet.

Structure of assets folder:

• assets
  • icons (all in git)
  • db
    • avonet.db
  • labels
    • bird_info.json
  • models
    • bird_model.onnx (the quantized onnx version of model20240824.pth)
    • ssd_mobilenet.onnx (the quantized version of pre-trained model from onnx modelzoo)

If you want to run batch classification for multiple photos, please check osea-cli.