X / Twitter
Version 1 almost done! Time to open source? https://t.co/hhSq6My74B
Everyone in education wants to help children develop a Love of Learning™
But nobody cares about the far more important and high-stakes task of helping a child develop a love of *effort*
Nobody, that is, except Montessori:
NariLabs Dia-1.6B by @_doyeob_ now on MLX 🚀
Get started:
> pip install -U mlx-audio
Model Card 👇🏾 https://t.co/IrrSu0mGR7 https://t.co/zwIavYjwbn
Code dropped: https://t.co/J6wjJsn9g6 https://t.co/SdIZvG0Glf
Moonshot just released Kimi-Audio on Hugging Face
SOTA on 10+ audio benchmarks: excels in Speech Recognition (LibriSpeech 1.28/2.42 WER), Audio Understanding (MMAU, VocalSound), and Conversation (VoiceBench) https://t.co/QXwFcUvkqj
Spotify just announced ViSMaP on Hugging Face
Unsupervised Hour-long Video Summarisation by Meta-Prompting https://t.co/lqTAqg3wJE
Step1X-Edit just dropped on Hugging Face
A Practical Framework for General Image Editing https://t.co/SnTB4AtynD
Breaking: @pleiasfr releases a new generation of small reasoning models for RAG and source synthesis. Pleias-RAG-350M and Pleias-RAG-1B come with built-in support for source citation, SOTA performance and an accuracy comparable to models ten times their size. https://t.co/vO8xqeVIxK
agent frameworks are overhead bc unnecessary abstraction. good for only fast prototyping. also, LLMs suck at working with cutting edge frameworks. a lot of agent dev boils down to software plumbing which LLMs excel at so just tool calling + programming primitives scales well. https://t.co/zgQSLRv7xD
Check out our recent #CVPR2025 paper AnyCam, a method for pose estimation in casual videos!
1️⃣ Can be directly trained on casual videos without the need for 3D annotation.
2️⃣ Based around a feed-forward transformer and light-weight refinement.
♦️ https://t.co/fB6pYn4uDJ https://t.co/mEPKe5SE2i
Introducing the Describe Anything Model (DAM), a powerful Multimodal LLM that generates detailed descriptions for user-specified regions in images or videos using points, boxes, scribbles, or masks.
Open-source code, models, demo, data, and benchmark at: https://t.co/2p3yM6mSUF https://t.co/cMw2GyxdTM
Alibaba just announced Uni3C on Hugging Face
Unifying Precisely 3D-Enhanced Camera and Human Motion Controls for Video Generation https://t.co/2tg4ofFphD
Introducing St4RTrack!🖖
Simultaneous 4D Reconstruction and Tracking in the world coordinate feed-forwardly, just by changing the meaning of two pointmaps!
https://t.co/ljcilL5CVJ https://t.co/nF217Ir03i
Mono3R: Exploiting Monocular Cues for Geometric 3D Reconstruction
Wenyu Li, Sidun Liu, Peng Qiao, Yong Dou
tl;dr: DUSt3R+MoGe+global Sim(3) alignment+monocular priors with monocular priors
https://t.co/MpngBfMAaY https://t.co/fZ5t1I9llv
New Anthropic research: AI values in the wild.
We want AI models to have well-aligned values. But how do we know what values they’re expressing in real-life conversations?
We studied hundreds of thousands of anonymized conversations to find out. https://t.co/EpOPguldZq
Editing reality with @runwayml Gen-4 https://t.co/RIL2KlcmGy
EDGS: Eliminating Densification for Efficient Convergence of 3DGS
Contributions:
• We show that initial triangulation based on 2D correspondences can replace the incremental refinement process, fundamentally changing how 3DGS models allocate resources.
• Our method reduces the path each Gaussian must travel in parameter space. Careful initialization not only accelerates convergence but also guides optimization toward a convergence point corresponding to lower reconstruction error and thus higher reconstruction quality.
• Our approach outperforms both speed-optimized and quality-focused state-of-the-art models while using only half the splats of standard 3DGS. By improving initialization rather than altering the optimization process, this method is compatible with other 3DGS acceleration techniques, making it a flexible enhancement to existing models.