LibreLGM0.4b

LGM, the Large Multi-View Gaussian Model ("LGM: Large Multi-View Gaussian Model for High-Resolution 3D Content Creation", ECCV 2024) by Jiaxiang Tang et al., repackaged for LibreYOLO. LGM reads four views of an object and returns it as 3D Gaussian splats; LibreYOLO generates the four views from one picture with ImageDream (LibreYOLO/LibreImageDream).

LibreYOLO runs it in plain PyTorch, without xformers or a rasterizer, so the same model runs on NVIDIA GPUs, Apple silicon (MPS) and CPU.

from libreyolo import LibreGen

model = LibreGen("lgm")          # downloads this repo and LibreImageDream
model("chair.jpg").save("chair.ply")

libreyolo generate source=chair.jpg model=lgm

This is a mirror, not a new model

The LGM weights are the authors'. Nothing was retrained, converted or re-encoded; model_fp16_fixrot.safetensors is byte-identical to upstream. It is the checkpoint upstream fine-tuned after its rotation-normalisation fix (April 2024), in fp16.

• Upstream repository: ashawkey/LGM
• Upstream revision: bc0097d24ffc5efc2e6a216edd579128f85f53ab
• Code: 3DTopia/LGM
• Paper: arXiv:2402.05054

File	Network
model_fp16_fixrot.safetensors	LGM (0.42B), upstream's big configuration
birefnet.safetensors	BiRefNet (Swin-L) background removal, MIT, by Peng Zheng et al.

Upstream's fp32 and pre-fix checkpoints are not mirrored. SHA-256 of every file is listed in NOTICE.

License

MIT, as published by the authors. See LICENSE and NOTICE. The ImageDream multi-view stage LibreYOLO runs in front of LGM is a separate repository under the OpenRAIL license.