How to train ldm use image as conditon?

[pokameng]

hello , thanks for you best works @rvencu @apolinario @hardmaru @TracelessLe @Enter-tainer I want to train a conditional model with ldm, using image as condition. Can you give me a advice about this? or share a example config for me??

[mia01]

Hi @pokameng were you able to find our how to do this? I have a similar problem

[sssimpleboy]

Hi @pokameng were you able to find our how to do this? I have a similar problem

Have you trained the model ?

[mia01]

@sssimpleboy I managed to take some huggingface code and amend it for my use case - I wanted to condition on layout conditions. I'm training the model but not very successful so far because I'm limited in resource. Although I'm using an A100 on colab pro - it requires more than 24 hours and not event completed an epoch ( I have 100k images). Not sure if anyone has any ideas on roughly how long such an endeavour could take

[sssimpleboy]

@sssimpleboy I managed to take some huggingface code and amend it for my use case - I wanted to condition on layout conditions. I'm training the model but not very successful so far because I'm limited in resource. Although I'm using an A100 on colab pro - it requires more than 24 hours and not event completed an epoch ( I have 100k images). Not sure if anyone has any ideas on roughly how long such an endeavour could take

Just now, I saw a file where the author said you can train the model with just 15-20 pictrues , but you should select them meticulously. Maybe you can reduce the number of pictrues.