Note Official evaluation benchmark with 1,500 distinct samples from 513 indoor scenes. Supplementary context-scaling experiments are documented separately. Release assets are in preparation.

Note 20,000 six-view training examples for BAGEL SFT. Training validation requires scene separation from the frozen benchmark. Dataset files are being prepared for release.

wangyanbo/MindView-BAGEL-SFT

Image-to-Image • Updated about 22 hours ago

Note BAGEL-SFT EMA weights for target-view generation. Inference requires the BAGEL base-model assets and compatible project runtime. Checkpoint files are being prepared for release.