search

raoyongming / CAL

[ICCV 2021] Counterfactual Attention Learning for Fine-Grained Visual Categorization and Re-identification
MIT License
145 stars 25 forks source link

How to train with single GPU? #14

Closed scutfrank closed 2 years ago

scutfrank commented 2 years ago

As above

raoyongming commented 2 years ago

Hi, the results of our models may be affected by the global batch size. You can just modify the number of GPUs to 1 but it may lead to a bit lower performance. Changing the learning rate or using gradient accumulation / fp16 can be helpful to alleviate the effects if you want to train the model with a single GPU.