Excited to share VidMap! #ECCV2026
We bring the temporal structure of SLAM into global SfM for accurate 3D reconstruction of long, unconstrained videos.
๐ Code: https://t.co/JSJFPk0rJK
๐ Paper: https://t.co/e5v8UHmggl
๐ฅ Video: https://t.co/xcD4hiOXuC
With @pesarlin & @mapo1
๐ [ECCV 2024 Oral] LGM: Large Multi-View Gaussian Model for High-Resolution 3D Content Creation.
๐น Gรฉnรฉration 3D haute-rรฉsolution
๐น Modรจle multi-vues avancรฉ
๐ https://t.co/Ih3DOv1HzG #Python
microsoft asia, coming out of nowhere, JUST dropped an image model on @huggingface ๐ซณ๐ผ๏ธ
Mage Flow is a 4B parameter DiT text-to-image and image editing model that can do images from 512x512 up to 4K
i just built a demo and am excited by the quality so far
demo: https://t.co/rae4UL39U0
model: https://t.co/jxuM7hGUdM
I did a quick and dirty weights only implementation of this. For the Accuracy Recovery Adapters, I have never been able to get them to work at 2bit, but this looks promising. Top row is normal 2 bit. Bottom row is OrbitQuant 2 bit. Going to test train an ARA with it
๐We released SeFi-Image, an open-sourced, more compute-efficient text-to-image foundation model.
โ Code and weights:
GitHub: https://t.co/SDyH2XkUUJ
Project Page: https://t.co/BdwfVw2GrS
Hugging Face: https://t.co/T25LpvcI1N
Can you decode AI latents to 4K images in under a second?
NVIDIA researchers introduce PiD, a pixel diffusion decoder that unifies decoding and upscaling into one fast step.
It converts 512x512 latents to 2048x2048 pixels in under 1 sec on a consumer RTX 5090 โ 6x faster than cascaded super-resolution pipelines, with better visual fidelity.
PiD: Fast and High-Resolution Latent Decoding with Pixel Diffusion
Paper: https://t.co/i5zf3w8BYq
Project: https://t.co/fvB8MWmtmW
Our report: https://t.co/IbCU91SVbH
๐ฌ #PapersAccepted by Jiqizhixin
Our simple rule: remove every part that seems to be removable. We starts with pixel space, the standard T5-L encoder, and a simple multimodal MM-JiT backbone with x-prediction.
Very exciting work with my amazing collaborators @Hope7Happiness, @Lyy_iiis, Kangyang Zhou, Linrui Ma, and Kaiming He!
All code, models, and full training recipe are open-sourced.
Blog Post: https://t.co/J1mmKH8tGI
Code: https://t.co/ConH73VHqn (JAX), https://t.co/tlxFdN3TLE (PyTorch).
๐ค Models: https://t.co/27MXZqhBNl