search

jolibrain / joliGEN

Generative AI Image Toolset with GANs and Diffusion for Real-World Applications
https://www.joligen.com
Other
236 stars 31 forks source link

DDPM inference anormal size #656

Closed YoannRandon closed 1 month ago

YoannRandon commented 3 months ago

Hello, i trained a palette model using the tutorials in "joligen.com/doc/". The training is now finished and i want to perform inference to look at my models performance ("visdom" didn't update with "nohup" command, I don't know why) So i used the command describ in the tutorial : image but i end up with the follow error : image I think 29Go is abnormally big, any thought about why the required memory is this high?

YoannRandon commented 3 months ago

more precisely, i trained the model on "Nuscenes" dataset, dimension = (1600*900) n015-2018-08-03-15-00-36+0800__CAM_FRONT__1533280329012460 Here, red square shows the bounding boxes in the associate ".txt" file

beniz commented 3 months ago

Hi, try with a smaller image to start with. Inference does not use amp/tf32 and/or torch.compile at the moment (training does with option --with_amp --with_tf32), but it could be added, saves half the memory in practice.