Question about the open-source timeline of the checkpoints? #2
Description
Hi!
Thank you for your great work! The results look very impressive.
Are there any plans to release the pre-trained model weights? We would be very grateful to test them and build upon your work.
I noticed that the provided training script (main.py) seems to load a local checkpoint for the LatentEventFusion model (which we believe is the MMCG) from a hardcoded path: /mnt/workspace/zhangziran/.../lef_model_checkpoint.pt
To help the community fully reproduce your paper's results, could you please release the pre-trained weights for both the lef_model (MMCG) and the fine-tuned SVD U-Net?
Uploading them to Hugging Face or as a GitHub release would be incredibly helpful.
Thank you!