Training Collaboration, Win-Win.
yuedajiong opened this issue · comments
I provide all "one-click-to-train" code.
Is anyone willing to provide computing power to collaborate on large-scale training together?
Costs:
computing:
4+ A100 (estimated,maybe more)
storage:
64T+ (estimated, maybe more)
for objvaserse download, and multi-view rendering; and image-features cache.
Profits:
model weights
all code, no limits
SOTA algorithms (various enhancements: image-feature-cache, fixed-and-generalized parameterd renderer, different priori source, priori hash/knn retrieval, equivariant-stereo-representation, loss-on-'pure'-2D(pure means without any camera-pose), ...)