search

guillaumejs2403 / TIME

Text-to-Image Models for Counterfactual Explanations: a black-box approach Official Code. WACV 2024
https://guillaumejs2403.github.io/projects/time.html
MIT License
7 stars 0 forks source link

Slow Textual Inversion #1

Open sidney1505 opened 1 month ago

sidney1505 commented 1 month ago

Hi,

thank you for this amazing work! The flexibility of this method is amazing.

However, textual inversion is very slow for me despite running it on a A100 it takes 10 seconds per step, while I would rather expect something like 10 steps per second as with normal diffusion training. Did you also encounter problems like that or have a guess why this happens?

Best regards, Sidney

guillaumejs2403 commented 1 month ago

Hi,

Sorry for the late response. How much time it is expected for it to train? Have you encountered data loading bottlenecks? Are you using any mini_batch_size?