search

nickgkan / butd_detr

Code for the ECCV22 paper "Bottom Up Top Down Detection Transformers for Language Grounding in Images and Point Clouds"
Other
74 stars 11 forks source link

Default Configuration for SR3D #22

Closed soham-joshi closed 1 year ago

soham-joshi commented 1 year ago

Hi @ayushjain1144 @nickgkan ,

A small query, what is the starting learning rate used for the experiments in Tables 6 and 7? The default configuration according to the code says --lr=1e-3, --lr_backbone=1e-4, and --text_encoder_lr=1e-5. But the script---train_test_cls.sh follows --lr=1e-4, --lr_backbone=1e-3, and --text_encoder_lr=1e-5.

Could you clarify the required configuration, please?

Thanks in advance!

ayushjain1144 commented 1 year ago

Hi @soham-joshi, train_test_cls.sh has the right configurations.

soham-joshi commented 1 year ago

Great, thanks for the prompt response!