search

mshukor / ViCHA

[BMVC22] Official Implementation of ViCHA: "Efficient Vision-Language Pretraining with Visual Concepts and Hierarchical Alignment"
MIT License
52 stars 1 forks source link

Batch size #2

Closed kimihailv closed 1 year ago

kimihailv commented 1 year ago

In addition, we train for 10 epochs (in contrast to 30 epochs [24, 43, 93]) with relatively small batch size of 256 (32 per GPU) using 4 GPUs.

Did you mean 64? 64 4 = 256, 32 4 = 128

mshukor commented 1 year ago

It's 32*4=128. Thanks