Junfei Xiao

Yutong Bai, Zeyu Wang, Junfei Xiao et al. · berkeley

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, ie, 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 a fine-tuned 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.

Junfei Xiao, Yutong Bai, Alan Yuille et al. · berkeley

Vision Transformer (ViT) has become one of the most popular neural architectures due to its great scalability, computational efficiency, and compelling performance in many vision tasks. However, ViT has shown inferior performance to Convolutional Neural Network (CNN) on medical tasks due to its data-hungry nature and the lack of annotated medical data. In this paper, we pre-train ViTs on 266,340 chest X-rays using Masked Autoencoders (MAE) which reconstruct missing pixels from a small part of each image. For comparison, CNNs are also pre-trained on the same 266,340 X-rays using advanced self-supervised methods (e.g., MoCo v2). The results show that our pre-trained ViT performs comparably (sometimes better) to the state-of-the-art CNN (DenseNet-121) for multi-label thorax disease classification. This performance is attributed to the strong recipes extracted from our empirical studies for pre-training and fine-tuning ViT. The pre-training recipe signifies that medical reconstruction requires a much smaller proportion of an image (10% vs. 25%) and a more moderate random resized crop range (0.5~1.0 vs. 0.2~1.0) compared with natural imaging. Furthermore, we remark that in-domain transfer learning is preferred whenever possible. The fine-tuning recipe discloses that layer-wise LR decay, RandAug magnitude, and DropPath rate are significant factors to consider. We hope that this study can direct future research on the application of Transformers to a larger variety of medical imaging tasks.

Junfei Xiao, Zhichao Xu, Shiyi Lan et al.

This report describes the winning solution to the Robust Vision Challenge (RVC) semantic segmentation track at ECCV 2022. Our method adopts the FAN-B-Hybrid model as the encoder and uses SegFormer as the segmentation framework. The model is trained on a composite dataset consisting of images from 9 datasets (ADE20K, Cityscapes, Mapillary Vistas, ScanNet, VIPER, WildDash 2, IDD, BDD, and COCO) with a simple dataset balancing strategy. All the original labels are projected to a 256-class unified label space, and the model is trained using a cross-entropy loss. Without significant hyperparameter tuning or any specific loss weighting, our solution ranks the first place on all the testing semantic segmentation benchmarks from multiple domains (ADE20K, Cityscapes, Mapillary Vistas, ScanNet, VIPER, and WildDash 2). The proposed method can serve as a strong baseline for the multi-domain segmentation task and benefit future works. Code will be available at https://github.com/lambert-x/RVC_Segmentation.

Jie Liu, Yixiao Zhang, Jie-Neng Chen et al.

An increasing number of public datasets have shown a marked impact on automated organ segmentation and tumor detection. However, due to the small size and partially labeled problem of each dataset, as well as a limited investigation of diverse types of tumors, the resulting models are often limited to segmenting specific organs/tumors and ignore the semantics of anatomical structures, nor can they be extended to novel domains. To address these issues, we propose the CLIP-Driven Universal Model, which incorporates text embedding learned from Contrastive Language-Image Pre-training (CLIP) to segmentation models. This CLIP-based label encoding captures anatomical relationships, enabling the model to learn a structured feature embedding and segment 25 organs and 6 types of tumors. The proposed model is developed from an assembly of 14 datasets, using a total of 3,410 CT scans for training and then evaluated on 6,162 external CT scans from 3 additional datasets. We rank first on the Medical Segmentation Decathlon (MSD) public leaderboard and achieve state-of-the-art results on Beyond The Cranial Vault (BTCV). Additionally, the Universal Model is computationally more efficient (6x faster) compared with dataset-specific models, generalized better to CT scans from varying sites, and shows stronger transfer learning performance on novel tasks.

Qixin Hu, Yixiong Chen, Junfei Xiao et al.

We demonstrate that AI models can accurately segment liver tumors without the need for manual annotation by using synthetic tumors in CT scans. Our synthetic tumors have two intriguing advantages: (I) realistic in shape and texture, which even medical professionals can confuse with real tumors; (II) effective for training AI models, which can perform liver tumor segmentation similarly to the model trained on real tumors -- this result is exciting because no existing work, using synthetic tumors only, has thus far reached a similar or even close performance to real tumors. This result also implies that manual efforts for annotating tumors voxel by voxel (which took years to create) can be significantly reduced in the future. Moreover, our synthetic tumors can automatically generate many examples of small (or even tiny) synthetic tumors and have the potential to improve the success rate of detecting small liver tumors, which is critical for detecting the early stages of cancer. In addition to enriching the training data, our synthesizing strategy also enables us to rigorously assess the AI robustness.

Qixin Hu, Junfei Xiao, Yixiong Chen et al.

We develop a novel strategy to generate synthetic tumors. Unlike existing works, the tumors generated by our strategy have two intriguing advantages: (1) realistic in shape and texture, which even medical professionals can confuse with real tumors; (2) effective for AI model training, which can perform liver tumor segmentation similarly to a model trained on real tumors - this result is unprecedented because no existing work, using synthetic tumors only, has thus far reached a similar or even close performance to the model trained on real tumors. This result also implies that manual efforts for developing per-voxel annotation of tumors (which took years to create) can be considerably reduced for training AI models in the future. Moreover, our synthetic tumors have the potential to improve the success rate of small tumor detection by automatically generating enormous examples of small (or tiny) synthetic tumors.

Siwei Yang, Longlong Jing, Junfei Xiao et al.

The weakly supervised instance segmentation is a challenging task. The existing methods typically use bounding boxes as supervision and optimize the network with a regularization loss term such as pairwise color affinity loss for instance segmentation. Through systematic analysis, we found that the commonly used pairwise affinity loss has two limitations: (1) it works with color affinity but leads to inferior performance with other modalities such as depth gradient, (2)the original affinity loss does not prevent trivial predictions as intended but actually accelerates this process due to the affinity loss term being symmetric. To overcome these two limitations, in this paper, we propose a novel asymmetric affinity loss which provides the penalty against the trivial prediction and generalizes well with affinity loss from different modalities. With the proposed asymmetric affinity loss, our method outperforms the state-of-the-art methods on the Cityscapes dataset and outperforms our baseline method by 3.5% in mask AP.

Sucheng Ren, Zeyu Wang, Hongru Zhu et al.

This paper enhances image-GPT (iGPT), one of the pioneering works that introduce autoregressive pretraining to predict the next pixels for visual representation learning. Two simple yet essential changes are made. First, we shift the prediction target from raw pixels to semantic tokens, enabling a higher-level understanding of visual content. Second, we supplement the autoregressive modeling by instructing the model to predict not only the next tokens but also the visible tokens. This pipeline is particularly effective when semantic tokens are encoded by discriminatively trained models, such as CLIP. We introduce this novel approach as D-iGPT. Extensive experiments showcase that D-iGPT excels as a strong learner of visual representations: A notable achievement is its compelling performance on the ImageNet-1K dataset -- by training on publicly available datasets, D-iGPT unprecedentedly achieves \textbf{90.0\%} top-1 accuracy with a vanilla ViT-H. Additionally, D-iGPT shows strong generalization on the downstream task. Code is available at https://github.com/OliverRensu/D-iGPT.

Junfei Xiao, Ziqi Zhou, Wenxuan Li et al.

This paper introduces ProLab, a novel approach using property-level label space for creating strong interpretable segmentation models. Instead of relying solely on category-specific annotations, ProLab uses descriptive properties grounded in common sense knowledge for supervising segmentation models. It is based on two core designs. First, we employ Large Language Models (LLMs) and carefully crafted prompts to generate descriptions of all involved categories that carry meaningful common sense knowledge and follow a structured format. Second, we introduce a description embedding model preserving semantic correlation across descriptions and then cluster them into a set of descriptive properties (e.g., 256) using K-Means. These properties are based on interpretable common sense knowledge consistent with theories of human recognition. We empirically show that our approach makes segmentation models perform stronger on five classic benchmarks (e.g., ADE20K, COCO-Stuff, Pascal Context, Cityscapes, and BDD). Our method also shows better scalability with extended training steps than category-level supervision. Our interpretable segmentation framework also emerges with the generalization ability to segment out-of-domain or unknown categories using only in-domain descriptive properties. Code is available at https://github.com/lambert-x/ProLab.

Xianhang Li, Haoqin Tu, Mude Hui et al.

Web-crawled image-text pairs are inherently noisy. Prior studies demonstrate that semantically aligning and enriching textual descriptions of these pairs can significantly enhance model training across various vision-language tasks, particularly text-to-image generation. However, large-scale investigations in this area remain predominantly closed-source. Our paper aims to bridge this community effort, leveraging the powerful and \textit{open-sourced} LLaMA-3, a GPT-4 level LLM. Our recaptioning pipeline is simple: first, we fine-tune a LLaMA-3-8B powered LLaVA-1.5 and then employ it to recaption 1.3 billion images from the DataComp-1B dataset. Our empirical results confirm that this enhanced dataset, Recap-DataComp-1B, offers substantial benefits in training advanced vision-language models. For discriminative models like CLIP, we observe enhanced zero-shot performance in cross-modal retrieval tasks. For generative models like text-to-image Diffusion Transformers, the generated images exhibit a significant improvement in alignment with users' text instructions, especially in following complex queries. Our project page is https://www.haqtu.me/Recap-Datacomp-1B/

Sucheng Ren, Xiaoke Huang, Xianhang Li et al.

This study presents Medical Vision Generalist (MVG), the first foundation model capable of handling various medical imaging tasks -- such as cross-modal synthesis, image segmentation, denoising, and inpainting -- within a unified image-to-image generation framework. Specifically, MVG employs an in-context generation strategy that standardizes the handling of inputs and outputs as images. By treating these tasks as an image generation process conditioned on prompt image-label pairs and input images, this approach enables a flexible unification of various tasks, even those spanning different modalities and datasets. To capitalize on both local and global context, we design a hybrid method combining masked image modeling with autoregressive training for conditional image generation. This hybrid approach yields the most robust performance across all involved medical imaging tasks. To rigorously evaluate MVG's capabilities, we curated the first comprehensive generalist medical vision benchmark, comprising 13 datasets and spanning four imaging modalities (CT, MRI, X-ray, and micro-ultrasound). Our results consistently establish MVG's superior performance, outperforming existing vision generalists, such as Painter and LVM. Furthermore, MVG exhibits strong scalability, with its performance demonstrably improving when trained on a more diverse set of tasks, and can be effectively adapted to unseen datasets with only minimal task-specific samples. The code is available at \url{https://github.com/OliverRensu/MVG}.

Junfei Xiao, Lequan Yu, Zongwei Zhou et al.

Batch normalization (BN) uniformly shifts and scales the activations based on the statistics of a batch of images. However, the intensity distribution of the background pixels often dominates the BN statistics because the background accounts for a large proportion of the entire image. This paper focuses on enhancing BN with the intensity distribution of foreground pixels, the one that really matters for image segmentation. We propose a new normalization strategy, named categorical normalization (CateNorm), to normalize the activations according to categorical statistics. The categorical statistics are obtained by dynamically modulating specific regions in an image that belong to the foreground. CateNorm demonstrates both precise and robust segmentation results across five public datasets obtained from different domains, covering complex and variable data distributions. It is attributable to the ability of CateNorm to capture domain-invariant information from multiple domains (institutions) of medical data. Code is available at https://github.com/lambert-x/CateNorm.

Taiming Lu, Tianmin Shu, Junfei Xiao et al.

Understanding, navigating, and exploring the 3D physical real world has long been a central challenge in the development of artificial intelligence. In this work, we take a step toward this goal by introducing GenEx, a system capable of planning complex embodied world exploration, guided by its generative imagination that forms priors (expectations) about the surrounding environments. GenEx generates an entire 3D-consistent imaginative environment from as little as a single RGB image, bringing it to life through panoramic video streams. Leveraging scalable 3D world data curated from Unreal Engine, our generative model is rounded in the physical world. It captures a continuous 360-degree environment with little effort, offering a boundless landscape for AI agents to explore and interact with. GenEx achieves high-quality world generation, robust loop consistency over long trajectories, and demonstrates strong 3D capabilities such as consistency and active 3D mapping. Powered by generative imagination of the world, GPT-assisted agents are equipped to perform complex embodied tasks, including both goal-agnostic exploration and goal-driven navigation. These agents utilize predictive expectation regarding unseen parts of the physical world to refine their beliefs, simulate different outcomes based on potential decisions, and make more informed choices. In summary, we demonstrate that GenEx provides a transformative platform for advancing embodied AI in imaginative spaces and brings potential for extending these capabilities to real-world exploration.

Shengqu Cai, Ceyuan Yang, Lvmin Zhang et al. · stanford

Long video generation is fundamentally a long context memory problem: models must retain and retrieve salient events across a long range without collapsing or drifting. However, scaling diffusion transformers to generate long-context videos is fundamentally limited by the quadratic cost of self-attention, which makes memory and computation intractable and difficult to optimize for long sequences. We recast long-context video generation as an internal information retrieval task and propose a simple, learnable sparse attention routing module, Mixture of Contexts (MoC), as an effective long-term memory retrieval engine. In MoC, each query dynamically selects a few informative chunks plus mandatory anchors (caption, local windows) to attend to, with causal routing that prevents loop closures. As we scale the data and gradually sparsify the routing, the model allocates compute to salient history, preserving identities, actions, and scenes over minutes of content. Efficiency follows as a byproduct of retrieval (near-linear scaling), which enables practical training and synthesis, and the emergence of memory and consistency at the scale of minutes.

Junfei Xiao, Ceyuan Yang, Lvmin Zhang et al. · stanford

We present Captain Cinema, a generation framework for short movie generation. Given a detailed textual description of a movie storyline, our approach firstly generates a sequence of keyframes that outline the entire narrative, which ensures long-range coherence in both the storyline and visual appearance (e.g., scenes and characters). We refer to this step as top-down keyframe planning. These keyframes then serve as conditioning signals for a video synthesis model, which supports long context learning, to produce the spatio-temporal dynamics between them. This step is referred to as bottom-up video synthesis. To support stable and efficient generation of multi-scene long narrative cinematic works, we introduce an interleaved training strategy for Multimodal Diffusion Transformers (MM-DiT), specifically adapted for long-context video data. Our model is trained on a specially curated cinematic dataset consisting of interleaved data pairs. Our experiments demonstrate that Captain Cinema performs favorably in the automated creation of visually coherent and narrative consistent short movies in high quality and efficiency. Project page: https://thecinema.ai

Junfei Xiao, Feng Cheng, Lu Qi et al.

Recent video generation models have shown promising results in producing high-quality video clips lasting several seconds. However, these models face challenges in generating long sequences that convey clear and informative events, limiting their ability to support coherent narrations. In this paper, we present a large-scale cooking video dataset designed to advance long-form narrative generation in the cooking domain. We validate the quality of our proposed dataset in terms of visual fidelity and textual caption accuracy using state-of-the-art Vision-Language Models (VLMs) and video generation models, respectively. We further introduce a Long Narrative Video Director to enhance both visual and semantic coherence in generated videos and emphasize the role of aligning visual embeddings to achieve improved overall video quality. Our method demonstrates substantial improvements in generating visually detailed and semantically aligned keyframes, supported by finetuning techniques that integrate text and image embeddings within the video generation process. Project page: https://videoauteur.github.io/

Yunfei Xie, Yinsong Ma, Shiyi Lan et al.

Developing reasoning capabilities in multimodal large language models (MLLMs) remains challenging. Motivated by literature suggesting that gameplay promotes transferable reasoning skills, we propose a novel post-training method, Visual Game Learning (ViGaL), where MLLMs develop generalizable reasoning skills through playing arcade-like games. Specifically, we show that training a 7B-parameter MLLM via reinforcement learning (RL) on simple games like Snake significantly enhances the downstream performance on multimodal math benchmarks like MathVista, on multi-discipline questions like MMMU and on 3D spatial reasoning benchmarks like VSI-Bench, without seeing any worked solutions, equations, or diagrams during RL. Remarkably, our model outperforms specialist models post-trained on benchmark-oriented multimodal reasoning data, while preserving the model's performance on general visual benchmarks, a challenge where specialist models often fall short. Our findings suggest that multimodal reasoning can emerge from gameplay, pointing to a promising strategy of designing surrogate tasks for RL post-training.

Junfei Xiao, Zheng Xu, Alan Yuille et al.

This paper demonstrates that a progressively aligned language model can effectively bridge frozen vision encoders and large language models (LLMs). While the fundamental architecture and pre-training methods of vision encoders and LLMs have been extensively studied, the architecture and training strategy of vision-language adapters vary significantly across recent works. Our research undertakes a thorough exploration of the state-of-the-art perceiver resampler architecture and builds a strong baseline. However, we observe that the vision-language alignment with perceiver resampler exhibits slow convergence and limited scalability with a lack of direct supervision. To address this issue, we propose PaLM2-VAdapter, employing a progressively aligned language model as the vision-language adapter. Compared to the strong baseline with perceiver resampler, our method empirically shows faster convergence, higher performance, and stronger scalability. Extensive experiments across various Visual Question Answering (VQA) and captioning tasks on both images and videos demonstrate that our model exhibits state-of-the-art visual understanding and multi-modal reasoning capabilities. Notably, our method achieves these advancements with 30~70% fewer parameters than the state-of-the-art large vision-language models, marking a significant efficiency improvement.

Tiezheng Zhang, Yitong Li, Yu-cheng Chou et al.

Building state-of-the-art Vision-Language Models (VLMs) with strong captioning capabilities typically necessitates training on billions of high-quality image-text pairs, requiring millions of GPU hours. This paper introduces the Vision-Language-Vision (VLV) auto-encoder framework, which strategically leverages key pretrained components: a vision encoder, the decoder of a Text-to-Image (T2I) diffusion model, and subsequently, a Large Language Model (LLM). Specifically, we establish an information bottleneck by regularizing the language representation space, achieved through freezing the pretrained T2I diffusion decoder. Our VLV pipeline effectively distills knowledge from the text-conditioned diffusion model using continuous embeddings, demonstrating comprehensive semantic understanding via high-quality reconstructions. Furthermore, by fine-tuning a pretrained LLM to decode the intermediate language representations into detailed descriptions, we construct a state-of-the-art (SoTA) captioner comparable to leading models like GPT-4o and Gemini 2.0 Flash. Our method demonstrates exceptional cost-efficiency and significantly reduces data requirements; by primarily utilizing single-modal images for training and maximizing the utility of existing pretrained models (image encoder, T2I diffusion model, and LLM), it circumvents the need for massive paired image-text datasets, keeping the total training expenditure under $1,000 USD.

Yu-Cheng Chou, Xingrui Wang, Yitong Li et al.

World engines aim to synthesize long, 3D-consistent videos that support interactive exploration of a scene under user-controlled camera motion. However, existing systems struggle under aggressive 6-DoF trajectories and complex outdoor layouts: they lose long-range geometric coherence, deviate from the target path, or collapse into overly conservative motion. To this end, we introduce Captain Safari, a pose-conditioned world engine that generates videos by retrieving from a persistent world memory. Given a camera path, our method maintains a dynamic local memory and uses a retriever to fetch pose-aligned world tokens, which then condition video generation along the trajectory. This design enables the model to maintain stable 3D structure while accurately executing challenging camera maneuvers. To evaluate this setting, we curate OpenSafari, a new in-the-wild FPV dataset containing high-dynamic drone videos with verified camera trajectories, constructed through a multi-stage geometric and kinematic validation pipeline. Across video quality, 3D consistency, and trajectory following, Captain Safari substantially outperforms state-of-the-art camera-controlled generators. It reduces MEt3R from 0.3703 to 0.3690, improves AUC@30 from 0.181 to 0.200, and yields substantially lower FVD than all camera-controlled baselines. More importantly, in a 50-participant, 5-way human study where annotators select the best result among five anonymized models, 67.6% of preferences favor our method across all axes. Our results demonstrate that pose-conditioned world memory is a powerful mechanism for long-horizon, controllable video generation and provide OpenSafari as a challenging new benchmark for future world-engine research.

Jiahao Wang, Luoxin Ye, TaiMing Lu et al.

Humans possess a remarkable ability to mentally explore and replay 3D environments they have previously experienced. Inspired by this mental process, we present EvoWorld: a world model that bridges panoramic video generation with evolving 3D memory to enable spatially consistent long-horizon exploration. Given a single panoramic image as input, EvoWorld first generates future video frames by leveraging a video generator with fine-grained view control, then evolves the scene's 3D reconstruction using a feedforward plug-and-play transformer, and finally synthesizes futures by conditioning on geometric reprojections from this evolving explicit 3D memory. Unlike prior state-of-the-arts that synthesize videos only, our key insight lies in exploiting this evolving 3D reconstruction as explicit spatial guidance for the video generation process, projecting the reconstructed geometry onto target viewpoints to provide rich spatial cues that significantly enhance both visual realism and geometric consistency. To evaluate long-range exploration capabilities, we introduce the first comprehensive benchmark spanning synthetic outdoor environments, Habitat indoor scenes, and challenging real-world scenarios, with particular emphasis on loop-closure detection and spatial coherence over extended trajectories. Extensive experiments demonstrate that our evolving 3D memory substantially improves visual fidelity and maintains spatial scene coherence compared to existing approaches, representing a significant advance toward long-horizon spatially consistent world modeling.

Junfei Xiao, Longlong Jing, Lin Zhang et al.

Semi-supervised video action recognition tends to enable deep neural networks to achieve remarkable performance even with very limited labeled data. However, existing methods are mainly transferred from current image-based methods (e.g., FixMatch). Without specifically utilizing the temporal dynamics and inherent multimodal attributes, their results could be suboptimal. To better leverage the encoded temporal information in videos, we introduce temporal gradient as an additional modality for more attentive feature extraction in this paper. To be specific, our method explicitly distills the fine-grained motion representations from temporal gradient (TG) and imposes consistency across different modalities (i.e., RGB and TG). The performance of semi-supervised action recognition is significantly improved without additional computation or parameters during inference. Our method achieves the state-of-the-art performance on three video action recognition benchmarks (i.e., Kinetics-400, UCF-101, and HMDB-51) under several typical semi-supervised settings (i.e., different ratios of labeled data).