Исключительно в оброзовательных целях
task is segmentation/object detection from drone's images on 4 classes: plastic, metall, wood, fisher's net
dataset links:
- original dataset: https://drive.google.com/file/d/10ZLtz6wcCRVWdxThjIfhEC-exJQGsrk0/view?usp=sharing (20 hand labelled images)
- hand labelled by our team: https://drive.google.com/file/d/11jUUARd8n2PwbI0Q15S2nrwSXnVBkkon/view?usp=sharing (50 images with metall barrels)
- photoshopped synthetic images: https://drive.google.com/file/d/1_oVzunUcsskHL_e-1yINgoynA9BLlk5C/view?usp=sharing (20 images with new/unseen types of litter)
trained models weights:
- detectron2: https://drive.google.com/file/d/1Tba6SY12osGkZL9oWaryN3MejNFshqcj/view?usp=sharing
- yolov5_l6(1280x1280px version): https://drive.google.com/file/d/1bhnIQ2JOjZRbddMwTkRAlkHrGG3W2s60/view?usp=sharing