Model Card for OpenLRM V1.1

Overview

• This model card is for the OpenLRM project, which is an open-source implementation of the paper LRM.
• Information contained in this model card corresponds to Version 1.1.

Model Details

• Training data

  Model	Training Data
  openlrm-obj-small-1.1	Objaverse
  openlrm-obj-base-1.1	Objaverse
  openlrm-obj-large-1.1	Objaverse
  openlrm-mix-small-1.1	Objaverse + MVImgNet
  openlrm-mix-base-1.1	Objaverse + MVImgNet
  openlrm-mix-large-1.1	Objaverse + MVImgNet

• Model architecture (version==1.1)

  Type	Layers	Feat. Dim	Attn. Heads	Triplane Dim.	Input Res.	Image Encoder	Size
  small	12	512	8	32	224	dinov2_vits14_reg	446M
  base	12	768	12	48	336	dinov2_vitb14_reg	1.04G
  large	16	1024	16	80	448	dinov2_vitb14_reg	1.81G

• Training settings

  Type	Rend. Res.	Rend. Patch	Ray Samples
  small	192	64	96
  base	288	96	96
  large	384	128	128

Notable Differences from the Original Paper

• We do not use the deferred back-propagation technique in the original paper.
• We used random background colors during training.
• The image encoder is based on the DINOv2 model with register tokens.
• The triplane decoder contains 4 layers in our implementation.

License

• The model weights are released under the Creative Commons Attribution-NonCommercial 4.0 International License.
• They are provided for research purposes only, and CANNOT be used commercially.

Disclaimer

Ethical Considerations

• This model should be used responsibly and ethically, and should not be used for malicious purposes.
• Users should be aware of potential biases in the training data.
• The model should not be used under the circumstances that could lead to harm or unfair treatment of individuals or groups.

Usage Considerations

• The model is provided "as is" without warranty of any kind.
• Users are responsible for ensuring that their use complies with all relevant laws and regulations.
• The developers and contributors of this model are not liable for any damages or losses arising from the use of this model.