search

FrozenBurning / Text2Light

[SIGGRAPH Asia 2022] Text2Light: Zero-Shot Text-Driven HDR Panorama Generation
https://frozenburning.github.io/projects/text2light/
Other
577 stars 46 forks source link

Retraining #14

Closed taginterco closed 1 year ago

taginterco commented 1 year ago

Where can I retrain the model with my own dataset? And what are the configurations for the model to read the dataset that I can provide

FrozenBurning commented 1 year ago

Thanks for your interest in our work. We have provided training details in the README. By training our model from scratch, we assume your dataset is stored in a folder containing lots of HDR panoramas (in .hdr/.exr format). Then, you can follow the instruction in README step by step to prepare training data.

taginterco commented 1 year ago

do the HDRI panoramas have to be a certain length and width? for example 2048x1024 or just 2:1 ratio @FrozenBurning

FrozenBurning commented 1 year ago

Recommend 2:1 ratio as it's the standard format of HDR panoramas. We trained on varied resolutions of HDR panoramas, and the preprocessing script will help you to handle all the things.