How to train with many objects

[mcondor10]

Hi! Currently I am working with cell instance segmentation, in which the datasets can contain up to 3000 objects per image. So I was wondering if there are tips or guidelines about how to train a model to detect that many objects. Is this achieved by changing the num_queries flag? Also, while doing evaluation, I see that it evaluates with a maximum of 100 detections. Is there a way to increase this value??

IoU metric: bbox
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.012
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.042

Thanks!

[kpbhat25]

Did you find any other model to train so many objects ??

[mcondor10]

Did you find any other model to train so many objects ??

Right now I'm trying with SAM (https://github.com/facebookresearch/segment-anything) but I haven't tried that many objects again yet.