Open bart1 opened 1 year ago
The point of interest becomes the graph. That graph is labeled (currently according to the POI label). The inference is done at this graph to determine the label of the graph. The inference of the POI is the inference of the graph.
https://github.com/point-cloud-radar/bird-cloud-gnn/blob/b04ad01abf6e30a2e94d032bf941b956c5b609f5/bird_cloud_gnn/gnn_model.py#L59 I think if I interpret this correct here we generate one label for the full graph, this might be tricky on the border of specific classes or for small objects in space. It would be good to also explore what happens if we label the specific point that was labeled.