salesforce / BLIP

PyTorch code for BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation
BSD 3-Clause "New" or "Revised" License
4.86k stars 648 forks source link

Validation time more than the training time #148

Open aman-cc opened 1 year ago

aman-cc commented 1 year ago

I'm training the BLIP captioning model on a custom dataset with 80:20 split (train:val). The training time for the epoch is ~6hrs while the val step takes ~42 hrs. Any idea why this is the case? Batch size seems to be the same for both steps.