search

ucbdrive / few-shot-object-detection

Implementations of few-shot object detection benchmarks
Apache License 2.0
1.1k stars 223 forks source link

About the test model's score threshold #128

Closed RuoyuChen10 closed 3 years ago

RuoyuChen10 commented 3 years ago

Thanks for your work, when I try to test the model by the command:

python -m tools.test_net --num-gpus 1 --config-file configs/PascalVOC-detection/split1/faster_rcnn_R_101_FPN_ft_all1_10shot.yaml --eval-all

I find that the threshold of the score when test is set as 0.05

Indicate that MODEL.ROI_HEADS.SCORE_THRESH_TEST=0.05

There will be lots of boxes, in few-shot learning, how much proposal boxes is used each image?

Or, we only visualize the image set a high threshold?

Looking forward to your reply.

RuoyuChen10 commented 3 years ago

Thanks for your work, when I try to test the model by the command:

python -m tools.test_net --num-gpus 1 --config-file configs/PascalVOC-detection/split1/faster_rcnn_R_101_FPN_ft_all1_10shot.yaml --eval-all

I find that the threshold of the score when test is set as 0.05

Indicate that MODEL.ROI_HEADS.SCORE_THRESH_TEST=0.05

There will be lots of boxes, in few-shot learning, how much proposal boxes is used each image?

Or, we only visualize the image set a high threshold?

Looking forward to your reply.

I am sorry for my poor knowledge in object detection, I forget the proposals will enter into NMS operation to remove the extra box.

thomasehuang commented 3 years ago

Exactly! No need to apologize, we are all here to learn :)

You can check the base config file for exact numbers regarding number of boxes if you're still curious.