Meta AI (FAIR)

Encoding expert music mixing conventions into language models can significantly elevate the quality of automatic music upmixing.

LeVJEPA achieves up to 20.8x less pretraining compute while surpassing the performance of leading video representation methods, reshaping the landscape of video-based learning.

VLMs can recognize when to abstain from making a decision, yet they fail to express this restraint, achieving a mere 0.292 on a new metric designed to measure this capability.

Early unification in multimodal training can prevent models from becoming overly reliant on language, unlocking new efficiencies in generative performance.

LeDXA reveals that self-supervised learning can unlock critical health insights from DXA scans, outperforming traditional metrics in predicting disease risk with far fewer resources.

Sparse microphone arrays can dramatically hinder acoustic mapping performance, but the right upsampling strategy can bridge the gap.

Achieving a 40% performance boost in robot control without the heavy computational costs of traditional vision-language models could revolutionize how we leverage visual representations in robotics.

Automated view scheduling in SceneFrom3D transforms the landscape of outdoor 3D scene generation, enabling unprecedented control over object appearance and geometry.

Gemma 4's unified architecture and reasoning mode enable it to outperform larger models in human-rated tasks while maintaining high efficiency.

Stochastic attention isn't just a regularizer; it fundamentally transforms how visual information is learned in VLMs, leading to more stable and reliable models.

LUNA achieves realistic 3D human animation from 2D inputs without the limitations of traditional skinning methods, enabling unprecedented flexibility and expressivity.

FiCA generates photorealistic avatars from a single image, achieving unprecedented visual quality and identity fidelity without the need for individual optimization.

Cross-modal prediction in MJEPA not only boosts performance but also reveals that a unified encoder can outperform traditional modality-specific approaches by leveraging shared information.

Relying on causal relationships rather than strong inductive biases, TDV achieves state-of-the-art performance in visual representation learning, challenging the status quo of self-supervised methods.

RepFusion reveals that multimodal large language models can dramatically enhance denoising in text-to-image systems, outperforming traditional denoising methods.

A new phase diagram reveals that cross-modal training can be actively harmful in certain contexts, guiding practitioners to choose the right approach before training.

Augmenting fMRI datasets with synthetic data can yield a staggering 68% boost in image retrieval accuracy, challenging traditional limits of brain decoding.

Forget task-specific fine-tuning – teaching VLMs basic geometry yields a +29% boost on spatial reasoning benchmarks.

Camera pose, largely ignored in video LLMs, unlocks significant gains in spatial reasoning and even improves general video QA when used as a lightweight supervisory signal.

Controllable 3D generation takes a leap forward with 3D-ReGen, a framework that leverages an initial 3D shape for tasks like enhancement and editing, outperforming existing methods.

Human motion generation gets a dose of reality: IAM shows that explicitly modeling body morphology and identity leads to more realistic and consistent movements.

Current egocentric video benchmarks miss the mark: EgoEverything uses human gaze to create questions that actually reflect how people behave, not just what they see.

A million videos with paired depth, camera pose, and 3D point tracks could unlock a new wave of 3D-aware video models.

Training 3D avatar diffusion models on millions of in-the-wild videos is now possible, thanks to a clever 3D tokenization and visibility-aware training strategy that overcomes partial observability.

Forget scaling laws: a large VLM strategically paired with a smaller model's reasoning tokens can rival the performance of a much larger, monolithic model.

Music-grounded video editing can now produce significantly more coherent timelines thanks to a novel global-local coordination mechanism that resolves cross-segment conflicts.

Unlocking the secrets of viral video ads: a new MLLM framework reveals which initial moments hook viewers and drive conversions.

By surgically intervening in MLLM decoding, this work cuts hallucination rates without sacrificing descriptive quality, a feat prior methods struggled to achieve.

Edit the bassline, drums, or other instruments of any song with this new open-source multi-stem music generation model.