This talk surveys my research agenda on advancing general-purpose visual intelligence, moving AI beyond static recognition toward active reasoning and embodied action. A central challenge is enabling AI systems to generalize reliably in low-data and long-tail regimes. I address this by combining multimodal representation learning with agentic reasoning frameworks such as PyVision, which equips vision models to dynamically generate tools for deliberate problem-solving, and ViGaL, which leverages gameplay to instill transferable cognitive skills for reasoning under scarcity. These efforts chart a trajectory from representation and generation to interactive, embodied agents, re-imagining AI as an active collaborator capable of tool use, imagination, and purposeful engagement across both digital and physical environments.

We introduce Perception Encoder (PE), a family of state-of-the-art vision encoders for image and video understanding. Traditionally, vision encoders have relied on a variety of pretraining objectives, each excelling at different downstream tasks. Surprisingly, after scaling a carefully tuned image pretraining recipe and refining with a robust video data engine, we find that contrastive vision-language training alone can produce strong, general embeddings for all of these downstream tasks. There is only one caveat: these embeddings are hidden within the intermediate layers of the network. To draw them out, we introduce two alignment methods: language alignment for multimodal language modeling, and spatial alignment for dense prediction. Together, our PE family of models achieves state-of-the-art results on a wide variety of tasks, including zero-shot image and video classification and retrieval; document, image, and video Q&A; and spatial tasks such as detection, tracking, and depth estimation. We release our models, code, and novel dataset of synthetically and human-annotated videos: https://github.com/facebookresearch/perception_models

We demonstrate text as a strong cross-modal interface. Rather than relying on deep embeddings to connect image and language as the interface representation, our approach represents an image as text, from which we enjoy the interpretability and flexibility inherent to natural language. We employ an autoencoder that uses a pre-trained text-to-image diffusion model for decoding. The encoder is trained to transform an input image into text, which is then fed into the fixed text-to-image diffusion decoder to reconstruct the original input ─a process we term De-Diffusion. Experiments validate both the precision and comprehensiveness of De-Diffusion text representing images, such that it can be readily ingested by off-the-shelf text-to-image tools and LLMs for diverse multi-modal tasks. For example, a single De-Diffusion model can generalize to provide transferable prompts for different text-to-image tools, and also achieves a new state of the art on open-ended vision-language tasks by simply prompting large language models with few-shot examples. Project page: dediffusion.github.io.

This paper introduces an efficient strategy to transform Large Language Models (LLMs) into Multi-Modal Large Language Models. By conceptualizing this transformation as a domain adaptation process, \ie, transitioning from text understanding to embracing multiple modalities, we intriguingly note that, within each attention block, tuning LayerNorm suffices to yield strong performance. Moreover, when benchmarked against other tuning approaches like full parameter finetuning or LoRA, its benefits on efficiency are substantial. For example, when compared to LoRA on a 13B model scale, performance can be enhanced by an average of over 20\% across five multi-modal tasks, and meanwhile, results in a significant reduction of trainable parameters by 41.9\% and a decrease in GPU memory usage by 17.6\%. On top of this LayerNorm strategy, we showcase that selectively tuning only with conversational data can improve efficiency further. Beyond these empirical outcomes, we provide a comprehensive analysis to explore the role of LayerNorm in adapting LLMs to the multi-modal domain and improving the expressive power of the model.

This paper studies the potential of distilling knowledge from pre-trained models, especially Masked Autoencoders. Our approach is simple: in addition to optimizing the pixel reconstruction loss on masked inputs, we minimize the distance between the intermediate feature map of the teacher model and that of the student model. This design leads to a computationally efficient knowledge distillation framework, given 1) only a small visible subset of patches is used, and 2) the (cumbersome) teacher model only needs to be partially executed, i.e., forward propagate inputs through the first few layers, for obtaining intermediate feature maps. Compared to directly distilling fine-tuned models, distilling pre-trained models substantially improves downstream performance. For example, by distilling the knowledge from an MAE pre-trained ViT-L into a ViT-B, our method achieves 84.0% ImageNet top-1 accuracy, outperforming the baseline of directly distilling afinetuned ViT-L by 1.2%. More intriguingly, our method can robustly distill knowledge from teacher models even with extremely high masking ratios: e.g., with 95% masking ratio where merely TEN patches are visible during distillation, our ViT-B competitively attains a top-1 ImageNet accuracy of 83.6%; surprisingly, it can still secure 82.4% top-1 ImageNet accuracy by aggressively training with just FOUR visible patches (98% masking ratio). The code and models are publicly available at https://github.com/UCSC-VLAA/DMAE.

There has been a longstanding belief that generation can facilitate a true understanding of visual data. In line with this, we revisit generatively pre-training visual representations in light of recent interest in denoising diffusion models. While directly pre-training with diffusion models does not produce strong representations, we condition diffusion models on masked input and formulate diffusion models as masked autoencoders (DiffMAE). Our approach is capable of (i) serving as a strong initialization for downstream recognition tasks, (ii) conducting high-quality image inpainting, and (iii) being effortlessly extended to video where it produces state-of-the-art classification accuracy. We further perform a comprehensive study on the pros and cons of design choices and build connections between diffusion models and masked autoencoders. Project page.

Video-language pre-training is crucial for learning powerful multi-modal representation. However, it typically requires a massive amount of computation. In this paper, we develop SMAUG, an efficient pre-training framework for video-language models. The foundation component in SMAUG is masked autoencoders. Different from prior works which only mask textual inputs, our masking strategy considers both visual and textual modalities, providing a better cross-modal alignment and saving more pre-training costs. On top of that, we introduce a space-time token sparsification module, which leverages context information to further select only "important" spatial regions and temporal frames for pre-training. Coupling all these designs allows our method to enjoy both competitive performances on text-to-video retrieval and video question answering tasks, and much less pre-training costs by 1.9× or more. For example, our SMAUG only needs ~50 NVIDIA A6000 GPU hours for pre-training to attain competitive performances on these two video-language tasks across six popular benchmarks.

Modern hierarchical vision transformers have added several vision-specific components in the pursuit of supervised classification performance. While these components lead to effective accuracies and attractive FLOP counts, the added complexity actually makes these transformers slower than their vanilla ViT counterparts. In this paper, we argue that this additional bulk is unnecessary. By pretraining with a strong visual pretext task (MAE), we can strip out all the bells-and-whistles from a state-of-the-art multi-stage vision transformer without losing accuracy. In the process, we create Hiera, an extremely simple hierarchical vision transformer that is more accurate than previous models while being significantly faster both at inference and during training. We evaluate Hiera on a variety of tasks for image and video recognition. Our code and models are available at https://github.com/facebookresearch/hiera.

We present Masked Feature Prediction (MaskFeat) for self-supervised pre-training of video models. Our approach first randomly masks out a portion of the input sequence and then predicts the feature of the masked regions. We study five different types of features and find Histograms of Oriented Gradients (HOG), a hand-crafted feature descriptor, works particularly well in terms of both performance and efficiency. We observe that the local contrast normalization in HOG is essential for good results, which is in line with earlier work using HOG for visual recognition. Our approach can learn abundant visual knowledge and drive large-scale Transformer based models. Without using extra model weights or supervision, MaskFeat pretrained on unlabeled videos achieves unprecedented results of 86.7% with MViTv2-L on Kinetics-400, 88.3% on Kinetics 600, 80.4% on Kinetics-700, 38.8 mAP on AVA, and 75.0% on SSv2. MaskFeat further generalizes to image input, which can be interpreted as a video with a single frame and obtains competitive results on ImageN et.

Semi-supervised learning on class-imbalanced data, although a realistic problem, has been under studied. While existing semi-supervised learning (SSL) methods are known to perform poorly on minority classes, we find that they still generate high precision pseudo-labels on minority classes. By exploiting this property, in this work, we propose Class-Rebalancing Self-Training (CReST), a simple yet effective framework to improve existing SSL methods on class-imbalanced data. CReST iteratively retrains a baseline SSL model with a labeled set expanded by adding pseudolabeled samples from an unlabeled set, where pseudolabeled samples from minority classes are selected more frequently according to an estimated class distribution. We also propose a progressive distribution alignment to adaptively adjust the rebalancing strength dubbed CReST+. We show that CReST and CReST+ improve state-of-the-art SSL algorithms on various class-imbalanced datasets and consistently outperform other popular rebalancing methods. Code has been made available at https://github.com/google-research/crest.

Learning visual features from unlabeled image data is an important yet challenging task, which is often achieved by training a model on some annotation-free information. We consider spatial contexts, for which we solve so-called jigsaw puzzles, i.e., each image is cut into grids and then disordered, and the goal is to recover the correct configuration. Existing approaches formulated it as a classification task by defining a fixed mapping from a small subset of configurations to a class set, but these approaches ignore the underlying relationship between different configurations and also limit their applications to more complex scenarios. This paper presents a novel approach which applies to jigsaw puzzles with an arbitrary grid size and dimensionality. We provide a fundamental and generalized principle, that weaker cues are easier to be learned in an unsupervised manner and also transfer better. In the context of puzzle recognition, we use an iterative manner which, instead of solving the puzzle all at once, adjusts the order of the patches in each step until convergence. In each step, we combine both unary and binary features of each patch into a cost function judging the correctness of the current configuration. Our approach, by taking similarity between puzzles into consideration, enjoys a more efficient way of learning visual knowledge. We verify the effectiveness of our approach from two aspects. First, it solves arbitrarily complex puzzles, including high-dimensional puzzles, that prior methods are difficult to handle. Second, it serves as a reliable way of network initialization, which leads to better transfer performance in visual recognition tasks including classification, detection and segmentation.

In this paper, we address the problem of lowlight enhancement. Our key idea is to first calculate a global illumination estimation for the low-light input, then adjust the illumination under the guidance of the estimation and supplement the details using a concatenation with the original input. Considering that, we propose a GLobal illumination Aware and Detail-preserving Network (GLADNet). The input image is rescaled to a certain size and then put into an encoder-decoder network to generate global priori knowledge of the illumination. Based on the global prior and the original input image, a convolutional network is employed for detail reconstruction. For training GLADNet, we use a synthetic dataset generated from RAW images. Extensive experiments demonstrate the superiority of our method over other compared methods on the real low-light images captured in various conditions.