For small and occluded objects about mmyolo HOT 5 CLOSED

i tried this but i want to utilisé detected information for further process that’s why i need self attention thing from where i can use response map correctly. because my goal is to fuse two images information to get one.

To the best of my knowledge, you can try the following approaches to boost the performance of existing detectors.

1. Strong backbone;
2. Modify the FPN in the network to ensure extreme small objects can be detected.
3. Stronger data augmentation method (e,g. grid mask and so on).

You can also alter the network structures to add attention mechanisms.

from mmyolo.

@noreenanwar For dense object detection you can try mmdet's crowddet.

from mmyolo.

The thing is that I want to use detected object attention maps as I am just interested for detected instance

from mmyolo.

The thing is that I want to use detected object attention maps as I am just interested for detected instance

I think you can try demo/featmap_vis_demo.py which provides cam map about the detection object.

from mmyolo.

i tried this but i want to utilisé detected information for further process that’s why i need self attention thing from where i can use response map correctly. because my goal is to fuse two images information to get one.

from mmyolo.