Hello! Really appreciate your work -- the paper and model and approach are very innovative and I'm grateful for you releasing your training / test data.
I work on tree counting as well, and I took your training dataset and used a DETR object detector to convert the points to bounding boxes. Here are a few examples:
Would you be interested in this data? I can share it with you if it would help your research. It matches very closely with the keypoint data:
I was interested in comparing your keypoint identification approach with a state-of-the-art object detection approach, and got the following results
So far, a DINO DETR (https://github.com/open-mmlab/mmdetection/blob/main/configs/dino/README.md) with a Vit-L base (DinoV2 SSL layer from our publication (https://www.sciencedirect.com/science/article/pii/S003442572300439X)) does very well!
Hello! Really appreciate your work -- the paper and model and approach are very innovative and I'm grateful for you releasing your training / test data.
I work on tree counting as well, and I took your training dataset and used a DETR object detector to convert the points to bounding boxes. Here are a few examples:
Would you be interested in this data? I can share it with you if it would help your research. It matches very closely with the keypoint data:
I was interested in comparing your keypoint identification approach with a state-of-the-art object detection approach, and got the following results
So far, a DINO DETR (https://github.com/open-mmlab/mmdetection/blob/main/configs/dino/README.md) with a Vit-L base (DinoV2 SSL layer from our publication (https://www.sciencedirect.com/science/article/pii/S003442572300439X)) does very well!