Bug in regression for depth/normal prediction using infer.py

[FantasticOven2]

Hello author,

Thanks for this amazing work!

When I try to run bash infer.sh with MODE="regression, I got the following error:

[b-burton]

Hey @FantasticOven2 I had the same issue as you. It is caused when trying to use the regression pipeline but you are inputting the generative model. You can see from the first line in your output you are loading the generative model "INFO:root: Successfully loading pipline from jingheya/lotus-depth-g-v1-0"

To fix this issue change:
export CHECKPOINT_DIR="jingheya/lotus-depth-g-v1-0"
To
export CHECKPOINT_DIR="jingheya/lotus-depth-d-v1-0"

Or if you are doing command line calls change the model flag from:

--pretrained_model_name_or_path="jingheya/lotus-depth-g-v1-0"
To
--pretrained_model_name_or_path="jingheya/lotus-depth-d-v1-0"

Keep in mind if you are doing normal estimation it is also a different model you need use than if you are doing depth estimation. The list of models available (and their paths) can be found here: https://github.com/EnVision-Research/Lotus?tab=readme-ov-file#choose-your-model

FYI the different model pipelines are quite large (>3GB)