Zhiquan-Wen / TDS

Pytorch Implementation for "Test-Time Model Adaptation for Visual Question Answering with Debiased Self-Supervisions" (IEEE TMM 2023).
BSD 3-Clause "New" or "Revised" License
3 stars 1 forks source link

可否提供一下LXMERT在VQACP-V1上的learning rate? #5

Closed LemonQC closed 1 year ago

LemonQC commented 1 year ago

目前复现的结果相比于论文低3个点,在改用lr=0.00002的时候,效果更差 image

Zhiquan-Wen commented 1 year ago

emm,你的learning rate错了吧,我设置的是0.0002,你这个低了一个数量级

LemonQC commented 1 year ago

好的,我尝试一下

LemonQC commented 1 year ago

基本复现