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

Do we need self.text_encoder.eval() #8

Closed Hiusam closed 1 year ago

Hiusam commented 1 year ago

Hi, it seems that in the code you fix the text_encoder by setting the requires_grad of its parameters to false. Do we need self.text_encoder.eval()? We have dropout in the text_encoder.

nickgkan commented 1 year ago

Hi, you're technically correct that we need to set the text_encoder in eval mode. In practice though, it won't affect much the result after many training iterations.