search

prs-eth / Marigold

[CVPR 2024 - Oral, Best Paper Award Candidate] Marigold: Repurposing Diffusion-Based Image Generators for Monocular Depth Estimation
https://marigoldmonodepth.github.io
Apache License 2.0
2.25k stars 124 forks source link

Questions about LCM model training #103

Closed luochunfen closed 1 month ago

luochunfen commented 2 months ago

Hi! I'm trying to train the LCM model. However, the model I trained is worse than the LCM model you released. Could you give more training details about LCM distillation ? Such as EMA decay, learning rate, training steps, optimizer. Is there any difference with the implementation of the official LCM (https://github.com/luosiallen/latent-consistency-model.git). Thank you for your help!

toshas commented 1 month ago

Hey thanks for your interest, we are preparing an LCM drop (paper, code), follow the repo/us on social media to learn when it is out

luochunfen commented 1 month ago

Hey thanks for your interest, we are preparing an LCM drop (paper, code), follow the repo/us on social media to learn when it is out

OK. Thank you for your reply.

omrastogi commented 1 month ago

Hi, any ideas when you'll be able to release the LCM drop?