Here we present baseline models we created while working with Baikal zooplankton images. Notebooks are accompanied by our second data repository.
Medium article with description of algorithms and architecture of all algorithms.
- These baseline notebooks represent first step towards high quality recognition in open world.
- Scheduled update of notebooks as well as segmentation models is planned.
- Purpose of models is to get along with open world classification and segmentation as well as with computer vision for micro world.
- We cannot guarantee reproducibility with new versions of dataset due to nature of solution and data scope.
Source dataset file data.json mentioned in notebooks is available here.
- Object segmentation notebook
- Embedding baseline notebook
- KNN shallow-dimension reasoning notebook
All models are released in unified ONNX format.
| Model | Link |
|---|---|
| Segmentation model | https://storage.yandexcloud.net/modelstorage/public/segmentator.onnx |
| Embedding model | https://storage.yandexcloud.net/modelstorage/public/embedder.onnx |
First release version of notebooks.