Loading video...
Video Failed to Load
拡散Transformerを稠密予測へ転用「Marigold V2」 🔹Qwen-Image-Edit-2509を4-bit量子化+Rank-128 QLoRAで微調整 🔹深度・法線・アルベドなどを1ステップで推定 🔹単一GPUで数日の学習を実現 STeRとの動画への応用可能性も示唆。 情報元はリプ欄。
61,045 views • 20 days ago •via X (Twitter)
5 Comments

Sadao Tokuyama20 days ago
情報元ページ

SirBadfish19 days ago
Tested this earlier today on HuggingFace in the Space. By far the best in class. It works really, really well.

Fajar M Reza19 days ago
Quantized QLoRA could make dense prediction research far more reproducible on modest hardware.

Leo Phiêu Diêu19 days ago
@tokufxug Hi, can it create grey-scale depth ?

Seltop19 days ago
holy cook

