search

ChenhongyiYang / PPAL

[CVPR 2024] Plug and Play Active Learning for Object Detection
Apache License 2.0
68 stars 8 forks source link

Different LR in configs than what's mentioned in the paper. #16

Open Y-T-G opened 1 month ago

Y-T-G commented 1 month ago

Is there a reason why the LR In the configs are different from MI-AOD which is what PPAL paper said it used as reference? The paper said that the MI-AOD config was followed, which is an LR of 0.001 for RetinaNet on both VOC0712 and COCO dataset.

PPAL image

MIAOD image

For example, for VOC0712, it's 0.002: https://github.com/ChenhongyiYang/PPAL/blob/15875ed7a524675bc6daeba79b3716a0abca2b64/configs/voc_active_learning/al_train/retinanet_26e.py#L22

And for COCO, it's 0.01: https://github.com/ChenhongyiYang/PPAL/blob/15875ed7a524675bc6daeba79b3716a0abca2b64/configs/coco_active_learning/al_train/retinanet_26e.py#L22