This would help the government and policy makers in trying to mitigate the issue. We manually collected water trash dataset and annotated more than 16,000 objects in 5400 images. For baseline study, we trained seven object detection models including YOLO, SSD, Faster-RCNN, RetinaNet, PeleeNet and M2Det. Overall, M2Det outperforms all other methods as it uses multiscale and multi-layer features from variable sized objects. Next, we plan to incorporate motion to track the detected trash, detect micro-particles such as plant leaves and improve interference time to deploy system on Raspberry-Pi or TX2 in remote sites to generate the statics of floating trash.
0 Comments
Leave a Reply. |
Archives
August 2019
Categories
|