Alan L. Yuille

ICLR Conference 2025 Conference Paper

Autoregressive Pretraining with Mamba in Vision

• Sucheng Ren
• Xianhang Li
• Haoqin Tu
• Feng Wang
• Fangxun Shu
• Lei Zhang
• Jieru Mei
• Linjie Yang

The vision community has started to build with the recently developed state space model, Mamba, as the new backbone for a range of tasks. This paper shows that Mamba's visual capability can be significantly enhanced through autoregressive pretraining, a direction not previously explored. Efficiency-wise, the autoregressive nature can well capitalize on the Mamba's unidirectional recurrent structure, enabling faster overall training speed compared to other training strategies like mask modeling. Performance-wise, autoregressive pretraining equips the Mamba architecture with markedly higher accuracy over its supervised-trained counterparts and, more importantly, successfully unlocks its scaling potential to large and even huge model sizes. For example, with autoregressive pretraining, a base-size Mamba attains 83.2\% ImageNet accuracy, outperforming its supervised counterpart by 2.0\%; our huge-size Mamba, the largest Vision Mamba to date, attains 85.0\% ImageNet accuracy (85.5\% when finetuned with $384\times384$ inputs), notably surpassing all other Mamba variants in vision. The code is available at \url{https://github.com/OliverRensu/ARM}.

ICLR Conference 2025 Conference Paper

Compositional 4D Dynamic Scenes Understanding with Physics Priors for Video Question Answering

• Xingrui Wang
• Wufei Ma
• Angtian Wang
• Shuo Chen
• Adam Kortylewski
• Alan L. Yuille

For vision-language models (VLMs), understanding the dynamic properties of objects and their interactions in 3D scenes from videos is crucial for effective reasoning about high-level temporal and action semantics. Although humans are adept at understanding these properties by constructing 3D and temporal (4D) representations of the world, current video understanding models struggle to extract these dynamic semantics, arguably because these models use cross-frame reasoning without underlying knowledge of the 3D/4D scenes. In this work, we introduce **DynSuperCLEVR**, the first video question answering dataset that focuses on language understanding of the dynamic properties of 3D objects. We concentrate on three physical concepts—*velocity*, *acceleration*, and *collisions*—within 4D scenes. We further generate three types of questions, including factual queries, future predictions, and counterfactual reasoning that involve different aspects of reasoning on these 4D dynamic properties. To further demonstrate the importance of explicit scene representations in answering these 4D dynamics questions, we propose **NS-4DPhysics**, a **N**eural-**S**ymbolic VideoQA model integrating **Physics** prior for **4D** dynamic properties with explicit scene representation of videos. Instead of answering the questions directly from the video text input, our method first estimates the 4D world states with a 3D generative model powered by a physical prior, and then uses neural symbolic reasoning to answer the questions based on the 4D world states. Our evaluation on all three types of questions in DynSuperCLEVR shows that previous video question answering models and large multimodal models struggle with questions about 4D dynamics, while our NS-4DPhysics significantly outperforms previous state-of-the-art models.

ICML Conference 2025 Conference Paper

FlowAR: Scale-wise Autoregressive Image Generation Meets Flow Matching

• Sucheng Ren
• Qihang Yu
• Ju He
• Xiaohui Shen
• Alan L. Yuille
• Liang-Chieh Chen

Autoregressive (AR) modeling has achieved remarkable success in natural language processing by enabling models to generate text with coherence and contextual understanding through next token prediction. Recently, in image generation, VAR proposes scale-wise autoregressive modeling, which extends the next token prediction to the next scale prediction, preserving the 2D structure of images. However, VAR encounters two primary challenges: (1) its complex and rigid scale design limits generalization in next scale prediction, and (2) the generator’s dependence on a discrete tokenizer with the same complex scale structure restricts modularity and flexibility in updating the tokenizer. To address these limitations, we introduce FlowAR, a general next scale prediction method featuring a streamlined scale design, where each subsequent scale is simply double the previous one. This eliminates the need for VAR’s intricate multi-scale residual tokenizer and enables the use of any off-the-shelf Variational AutoEncoder (VAE). Our simplified design enhances generalization in next scale prediction and facilitates the integration of Flow Matching for high-quality image synthesis. We validate the effectiveness of FlowAR on the challenging ImageNet-256 benchmark, demonstrating superior generation performance compared to previous methods. Codes is available at https: //github. com/OliverRensu/FlowAR.

ICLR Conference 2025 Conference Paper

GenEx: Generating an Explorable World

• Taiming Lu
• Tianmin Shu
• Alan L. Yuille
• Daniel Khashabi
• Jieneng Chen

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 expectations about the surrounding environments. *GenEx* generates high-quality, continuous 360-degree virtual environments, achieving robust loop consistency and active 3D mapping over extended trajectories. Leveraging generative imagination, GPT-assisted agents can undertake complex embodied tasks, including goal-agnostic exploration and goal-driven navigation. Agents utilize imagined observations to update their beliefs, simulate potential outcomes, and enhance their decision-making. Training on the synthetic urban dataset *GenEx-DB* and evaluation on *GenEx-EQA* demonstrate that our approach significantly improves agents' planning capabilities, providing a transformative platform toward intelligent, imaginative embodied exploration.

ICML Conference 2025 Conference Paper

Scaling Laws in Patchification: An Image Is Worth 50, 176 Tokens And More

• Feng Wang 0047
• Yaodong Yu
• Wei Shao 0008
• Yuyin Zhou
• Alan L. Yuille
• Cihang Xie

Since the introduction of Vision Transformer (ViT), patchification has long been regarded as a common image pre-processing approach for plain visual architectures. By compressing the spatial size of images, this approach can effectively shorten the token sequence and reduce the computational cost of ViT-like plain architectures. In this work, we aim to thoroughly examine the information loss caused by this patchification-based compressive encoding paradigm and how it affects visual understanding. We conduct extensive patch size scaling experiments and excitedly observe an intriguing scaling law in patchification: the models can consistently benefit from decreased patch sizes and attain improved predictive performance, until it reaches the minimum patch size of 1*1, i. e. , pixel tokenization. This conclusion is broadly applicable across different vision tasks, various input scales, and diverse architectures such as ViT and the recent Mamba models. Moreover, as a by-product, we discover that with smaller patches, task-specific decoder heads become less critical for dense prediction. In the experiments, we successfully scale up the visual sequence to an exceptional length of 50, 176 tokens, achieving a competitive test accuracy of 84. 6% with a base-sized model on the ImageNet-1k benchmark. We hope this study can provide insights and theoretical foundations for future works of building non-compressive vision models.

ICLR Conference 2024 Conference Paper

Discovering Failure Modes of Text-guided Diffusion Models via Adversarial Search

• Qihao Liu
• Adam Kortylewski
• Yutong Bai
• Song Bai 0001
• Alan L. Yuille

Text-guided diffusion models (TDMs) are widely applied but can fail unexpectedly. Common failures include: _(i)_ natural-looking text prompts generating images with the wrong content, or _(ii)_ different random samples of the latent variables that generate vastly different, and even unrelated, outputs despite being conditioned on the same text prompt. In this work, we aim to study and understand the failure modes of TDMs in more detail. To achieve this, we propose SAGE, the first adversarial search method on TDMs that systematically explores the discrete prompt space and the high-dimensional latent space, to automatically discover undesirable behaviors and failure cases in image generation. We use image classifiers as surrogate loss functions during searching, and employ human inspections to validate the identified failures. For the first time, our method enables efficient exploration of both the discrete and intricate human language space and the challenging latent space, overcoming the gradient vanishing problem. Then, we demonstrate the effectiveness of SAGE on five widely used generative models and reveal four typical failure modes that have not been systematically studied before: (1) We find a variety of natural text prompts that generate images failing to capture the semantics of input texts. We further discuss the underlying causes and potential solutions based on the results. (2) We find regions in the latent space that lead to distorted images independent of the text prompt, suggesting that parts of the latent space are not well-structured. (3) We also find latent samples that result in natural-looking images unrelated to the text prompt, implying a possible misalignment between the latent and prompt spaces. (4) By appending a single adversarial token embedding to any input prompts, we can generate a variety of specified target objects, with minimal impact on CLIP scores, demonstrating the fragility of language representations.

ICLR Conference 2024 Conference Paper

Generating Images with 3D Annotations Using Diffusion Models

• Wufei Ma
• Qihao Liu
• Jiahao Wang 0001
• Angtian Wang
• Xiaoding Yuan
• Yi Zhang 0099
• Zihao Xiao 0001
• Guofeng Zhang 0020

Diffusion models have emerged as a powerful generative method, capable of producing stunning photo-realistic images from natural language descriptions. However, these models lack explicit control over the 3D structure in the generated images. Consequently, this hinders our ability to obtain detailed 3D annotations for the generated images or to craft instances with specific poses and distances. In this paper, we propose 3D Diffusion Style Transfer (3D-DST), which incorporates 3D geometry control into diffusion models. Our method exploits ControlNet, which extends diffusion models by using visual prompts in addition to text prompts. We generate images of the 3D objects taken from 3D shape repositories~(e.g., ShapeNet and Objaverse), render them from a variety of poses and viewing directions, compute the edge maps of the rendered images, and use these edge maps as visual prompts to generate realistic images. With explicit 3D geometry control, we can easily change the 3D structures of the objects in the generated images and obtain ground-truth 3D annotations automatically. This allows us to improve a wide range of vision tasks, e.g., classification and 3D pose estimation, in both in-distribution (ID) and out-of-distribution (OOD) settings. We demonstrate the effectiveness of our method through extensive experiments on ImageNet-100/200, ImageNet-R, PASCAL3D+, ObjectNet3D, and OOD-CV. The results show that our method significantly outperforms existing methods, e.g., 3.8 percentage points on ImageNet-100 using DeiT-B. Our code is available at <https://ccvl.jhu.edu/3D-DST/>

ICLR Conference 2024 Conference Paper

How Well Do Supervised 3D Models Transfer to Medical Imaging Tasks?

• Wenxuan Li
• Alan L. Yuille
• Zongwei Zhou

The pre-training and fine-tuning paradigm has become prominent in transfer learning. For example, if the model is pre-trained on ImageNet and then fine-tuned to PASCAL, it can significantly outperform that trained on PASCAL from scratch. While ImageNet pre-training has shown enormous success, it is formed in 2D, and the learned features are for classification tasks; when transferring to more diverse tasks, like 3D image segmentation, its performance is inevitably compromised due to the deviation from the original ImageNet context. A significant challenge lies in the lack of large, annotated 3D datasets rivaling the scale of ImageNet for model pre-training. To overcome this challenge, we make two contributions. Firstly, we construct AbdomenAtlas 1.1 that comprises **9,262** three-dimensional computed tomography (CT) volumes with high-quality, per-voxel annotations of 25 anatomical structures and pseudo annotations of seven tumor types. Secondly, we develop a suite of models that are pre-trained on our AbdomenAtlas 1.1 for transfer learning. Our preliminary analyses indicate that the model trained only with 21 CT volumes, 672 masks, and 40 GPU hours has a transfer learning ability similar to the model trained with 5,050 (unlabeled) CT volumes and 1,152 GPU hours. More importantly, the transfer learning ability of supervised models can further scale up with larger annotated datasets, achieving significantly better performance than preexisting pre-trained models, irrespective of their pre-training methodologies or data sources. We hope this study can facilitate collective efforts in constructing larger 3D medical datasets and more releases of supervised pre-trained models.

ICML Conference 2024 Conference Paper

Rejuvenating image-GPT as Strong Visual Representation Learners

• Sucheng Ren
• Zeyu Wang 0008
• Hongru Zhu
• Junfei Xiao
• Alan L. Yuille
• Cihang Xie

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 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.

ICLR Conference 2024 Conference Paper

Source-Free and Image-Only Unsupervised Domain Adaptation for Category Level Object Pose Estimation

• Prakhar Kaushik
• Aayush Mishra
• Adam Kortylewski
• Alan L. Yuille

We consider the problem of source-free unsupervised category-level 3D pose estimation from only RGB images to an non-annotated and unlabelled target domain without any access to source domain data or annotations during adaptation. Collecting and annotating real world 3D data and corresponding images is laborious, expensive yet unavoidable process since even 3D pose domain adaptation methods require 3D data in the target domain. We introduce a method which is capable of adapting to a nuisance ridden target domain without any 3D data or annotations. We represent object categories as simple cuboid meshes, and harness a generative model of neural feature activations modeled as a von Mises Fisher distribution at each mesh vertex learnt using differential rendering. We focus on individual mesh vertex features and iteratively update them based on their proximity to corresponding features in the target domain. Our key insight stems from the observation that specific object subparts remain stable across out-of-domain (OOD) scenarios, enabling strategic utilization of these invariant subcomponents for effective model updates. Our model is then trained in an EM fashion alternating between updating the vertex features and feature extractor. We show that our method simulates fine-tuning on a global-pseudo labelled dataset under mild assumptions which converges to the target domain asymptotically. Through extensive empirical validation, we demonstrate the potency of our simple approach in addressing the domain shift challenge and significantly enhancing pose estimation accuracy. By accentuating robust and less changed object subcomponents, our framework contributes to the evolution of UDA techniques in the context of 3D pose estimation using only images from the target domain.

NeurIPS Conference 2024 Conference Paper

Touchstone Benchmark: Are We on the Right Way for Evaluating AI Algorithms for Medical Segmentation?

• Pedro R. Bassi
• Wenxuan Li
• Yucheng Tang
• Fabian Isensee
• Zifu Wang
• Jieneng Chen
• Yu-Cheng Chou
• Saikat Roy

How can we test AI performance? This question seems trivial, but it isn't. Standard benchmarks often have problems such as in-distribution and small-size test sets, oversimplified metrics, unfair comparisons, and short-term outcome pressure. As a consequence, good performance on standard benchmarks does not guarantee success in real-world scenarios. To address these problems, we present Touchstone, a large-scale collaborative segmentation benchmark of 9 types of abdominal organs. This benchmark is based on 5, 195 training CT scans from 76 hospitals around the world and 5, 903 testing CT scans from 11 additional hospitals. This diverse test set enhances the statistical significance of benchmark results and rigorously evaluates AI algorithms across various out-of-distribution scenarios. We invited 14 inventors of 19 AI algorithms to train their algorithms, while our team, as a third party, independently evaluated these algorithms on three test sets. In addition, we also evaluated pre-existing AI frameworks---which, differing from algorithms, are more flexible and can support different algorithms—including MONAI from NVIDIA, nnU-Net from DKFZ, and numerous other open-source frameworks. We are committed to expanding this benchmark to encourage more innovation of AI algorithms for the medical domain.

NeurIPS Conference 2023 Conference Paper

3D-Aware Visual Question Answering about Parts, Poses and Occlusions

• Xingrui Wang
• Wufei Ma
• Zhuowan Li
• Adam Kortylewski
• Alan L. Yuille

Despite rapid progress in Visual question answering (\textit{VQA}), existing datasets and models mainly focus on testing reasoning in 2D. However, it is important that VQA models also understand the 3D structure of visual scenes, for example to support tasks like navigation or manipulation. This includes an understanding of the 3D object pose, their parts and occlusions. In this work, we introduce the task of 3D-aware VQA, which focuses on challenging questions that require a compositional reasoning over the 3D structure of visual scenes. We address 3D-aware VQA from both the dataset and the model perspective. First, we introduce Super-CLEVR-3D, a compositional reasoning dataset that contains questions about object parts, their 3D poses, and occlusions. Second, we propose PO3D-VQA, a 3D-aware VQA model that marries two powerful ideas: probabilistic neural symbolic program execution for reasoning and deep neural networks with 3D generative representations of objects for robust visual recognition. Our experimental results show our model PO3D-VQA outperforms existing methods significantly, but we still observe a significant performance gap compared to 2D VQA benchmarks, indicating that 3D-aware VQA remains an important open research area.

NeurIPS Conference 2023 Conference Paper

AbdomenAtlas-8K: Annotating 8,000 CT Volumes for Multi-Organ Segmentation in Three Weeks

• Chongyu Qu
• Tiezheng Zhang
• Hualin Qiao
• Jie Liu
• Yucheng Tang
• Alan L. Yuille
• Zongwei Zhou

Annotating medical images, particularly for organ segmentation, is laborious and time-consuming. For example, annotating an abdominal organ requires an estimated rate of 30-60 minutes per CT volume based on the expertise of an annotator and the size, visibility, and complexity of the organ. Therefore, publicly available datasets for multi-organ segmentation are often limited in data size and organ diversity. This paper proposes an active learning procedure to expedite the annotation process for organ segmentation and creates the largest multi-organ dataset (by far) with the spleen, liver, kidneys, stomach, gallbladder, pancreas, aorta, and IVC annotated in 8, 448 CT volumes, equating to 3. 2 million slices. The conventional annotation methods would take an experienced annotator up to 1, 600 weeks (or roughly 30. 8 years) to complete this task. In contrast, our annotation procedure has accomplished this task in three weeks (based on an 8-hour workday, five days a week) while maintaining a similar or even better annotation quality. This achievement is attributed to three unique properties of our method: (1) label bias reduction using multiple pre-trained segmentation models, (2) effective error detection in the model predictions, and (3) attention guidance for annotators to make corrections on the most salient errors. Furthermore, we summarize the taxonomy of common errors made by AI algorithms and annotators. This allows for continuous improvement of AI and annotations, significantly reducing the annotation costs required to create large-scale datasets for a wider variety of medical imaging tasks. Code and dataset are available at https: //github. com/MrGiovanni/AbdomenAtlas

ICLR Conference 2023 Conference Paper

MOAT: Alternating Mobile Convolution and Attention Brings Strong Vision Models

• Chenglin Yang
• Siyuan Qiao
• Qihang Yu
• Xiaoding Yuan
• Yukun Zhu
• Alan L. Yuille
• Hartwig Adam
• Liang-Chieh Chen

This paper presents MOAT, a family of neural networks that build on top of MObile convolution (i.e., inverted residual blocks) and ATtention. Unlike the current works that stack separate mobile convolution and transformer blocks, we effectively merge them into a MOAT block. Starting with a standard Transformer block, we replace its multi-layer perceptron with a mobile convolution block, and further reorder it before the self-attention operation. The mobile convolution block not only enhances the network representation capacity, but also produces better downsampled features. Our conceptually simple MOAT networks are surprisingly effective, achieving 89.1% / 81.5% top-1 accuracy on ImageNet-1K / ImageNet-1K-V2 with ImageNet-22K pretraining. Additionally, MOAT can be seamlessly applied to downstream tasks that require large resolution inputs by simply converting the global attention to window attention. Thanks to the mobile convolution that effectively exchanges local information between pixels (and thus cross-windows), MOAT does not need the extra window-shifting mechanism. As a result, on COCO object detection, MOAT achieves 59.2% AP$^{\text{box}}$ with 227M model parameters (single-scale inference, and hard NMS), and on ADE20K semantic segmentation, MOAT attains 57.6% mIoU with 496M model parameters (single-scale inference). Finally, the tiny-MOAT family, obtained by simply reducing the channel sizes, also surprisingly outperforms several mobile-specific transformer-based models on ImageNet. The tiny-MOAT family is also benchmarked on downstream tasks, serving as a baseline for the community. We hope our simple yet effective MOAT will inspire more seamless integration of convolution and self-attention. Code is publicly available.

ICLR Conference 2023 Conference Paper

VoGE: A Differentiable Volume Renderer using Gaussian Ellipsoids for Analysis-by-Synthesis

• Angtian Wang
• Peng Wang 0001
• Jian Sun
• Adam Kortylewski
• Alan L. Yuille

Differentiable rendering allows the application of computer graphics on vision tasks, e.g. object pose and shape fitting, via analysis-by-synthesis, where gradients at occluded regions are important when inverting the rendering process.To obtain those gradients, state-of-the-art (SoTA) differentiable renderers use rasterization to collect a set of nearest components for each pixel and aggregate them based on the viewing distance. In this paper, we propose VoGE, which uses ray tracing to capture nearest components with their volume density distributions on the rays and aggregates via integral of the volume densities based on Gaussian ellipsoids, which brings more efficient and stable gradients. To efficiently render via VoGE, we propose an approximate close-form solution for the volume density aggregation and a coarse-to-fine rendering strategy. Finally, we provide a CUDA implementation of VoGE, which gives a competitive rendering speed in comparison to PyTorch3D. Quantitative and qualitative experiment results show VoGE outperforms SoTA counterparts when applied to various vision tasks, e.g., object pose estimation, shape/texture fitting, and occlusion reasoning. The VoGE code is available at: https://github.com/Angtian/VoGE.

ICLR Conference 2023 Conference Paper

Which Layer is Learning Faster? A Systematic Exploration of Layer-wise Convergence Rate for Deep Neural Networks

• Yixiong Chen
• Alan L. Yuille
• Zongwei Zhou

The deeply hierarchical structures enable deep neural networks (DNNs) to fit extremely complex target functions. However, the complex interaction between layers also makes the learning process of a particular layer poorly understood. This work demonstrates that the shallower layers of DNNs tend to converge faster than the deeper layers. We call this phenomenon Layer Convergence Bias. We also uncover the fundamental reason behind this phenomenon: Flatter local minima of shallower layers make their gradients more stable and predictive, allowing for faster training. Another surprising result is that the shallower layers tend to learn the low-frequency components of the target function, while the deeper layers usually learn the high-frequency components. It is consistent with the recent discovery that DNNs learn lower frequency objects faster.

ICLR Conference 2022 Conference Paper

Fast AdvProp

• Jieru Mei
• Yucheng Han
• Yutong Bai
• Yixiao Zhang 0001
• Yingwei Li 0002
• Xianhang Li
• Alan L. Yuille
• Cihang Xie

Adversarial Propagation (AdvProp) is an effective way to improve recognition models, leveraging adversarial examples. Nonetheless, AdvProp suffers from the extremely slow training speed, mainly because: a) extra forward and backward passes are required for generating adversarial examples; b) both original samples and their adversarial counterparts are used for training (i.e., 2X data). In this paper, we introduce Fast AdvProp, which aggressively revamps AdvProp's costly training components, rendering the method nearly as cheap as the vanilla training. Specifically, our modifications in Fast AdvProp are guided by the hypothesis that disentangled learning with adversarial examples is the key for performance improvements, while other training recipes (e.g., paired clean and adversarial training samples, multi-step adversarial attackers) could be largely simplified. Our empirical results show that, compared to the vanilla training baseline, Fast AdvProp is able to further model performance on a spectrum of visual benchmarks, without incurring extra training cost. Additionally, our ablations find Fast AdvProp scales better if larger models are used, is compatible with existing data augmentation methods (i.e., Mixup and CutMix), and can be easily adapted to other recognition tasks like object detection. The code is available here: https://github.com/meijieru/fast_advprop.

ICLR Conference 2022 Conference Paper

Image BERT Pre-training with Online Tokenizer

• Jinghao Zhou
• Chen Wei 0005
• Huiyu Wang
• Wei Shen 0002
• Cihang Xie
• Alan L. Yuille
• Tao Kong

The success of language Transformers is primarily attributed to the pretext task of masked language modeling (MLM), where texts are first tokenized into semantically meaningful pieces. In this work, we study masked image modeling (MIM) and indicate the necessity and challenges of using a semantically meaningful visual tokenizer. We present a self-supervised framework iBOT that can perform masked prediction with an online tokenizer. Specifically, we perform self-distillation on masked patch tokens and take the teacher network as the online tokenizer, along with self-distillation on the class token to acquire visual semantics. The online tokenizer is jointly learnable with the MIM objective and dispenses with a multi-stage training pipeline where the tokenizer needs to be pre-trained beforehand. We show the prominence of iBOT by achieving an 82.3% linear probing accuracy and an 87.8% fine-tuning accuracy evaluated on ImageNet-1K. Beyond the state-of-the-art image classification results, we underline emerging local semantic patterns, which helps the models to obtain strong robustness against common corruptions and achieve leading results on dense downstream tasks, e.g., object detection, instance segmentation, and semantic segmentation.

NeurIPS Conference 2021 Conference Paper

Are Transformers more robust than CNNs?

• Yutong Bai
• Jieru Mei
• Alan L. Yuille
• Cihang Xie

Transformer emerges as a powerful tool for visual recognition. In addition to demonstrating competitive performance on a broad range of visual benchmarks, recent works also argue that Transformers are much more robust than Convolutions Neural Networks (CNNs). Nonetheless, surprisingly, we find these conclusions are drawn from unfair experimental settings, where Transformers and CNNs are compared at different scales and are applied with distinct training frameworks. In this paper, we aim to provide the first fair & in-depth comparisons between Transformers and CNNs, focusing on robustness evaluations. With our unified training setup, we first challenge the previous belief that Transformers outshine CNNs when measuring adversarial robustness. More surprisingly, we find CNNs can easily be as robust as Transformers on defending against adversarial attacks, if they properly adopt Transformers' training recipes. While regarding generalization on out-of-distribution samples, we show pre-training on (external) large-scale datasets is not a fundamental request for enabling Transformers to achieve better performance than CNNs. Moreover, our ablations suggest such stronger generalization is largely benefited by the Transformer's self-attention-like architectures per se, rather than by other training setups. We hope this work can help the community better understand and benchmark the robustness of Transformers and CNNs. The code and models are publicly available at: https: //github. com/ytongbai/ViTs-vs-CNNs.

ICLR Conference 2021 Conference Paper

CO2: Consistent Contrast for Unsupervised Visual Representation Learning

• Chen Wei 0005
• Huiyu Wang
• Wei Shen 0002
• Alan L. Yuille

Contrastive learning has recently been a core for unsupervised visual representation learning. Without human annotation, the common practice is to perform an instance discrimination task: Given a query image crop, label crops from the same image as positives, and crops from other randomly sampled images as negatives. An important limitation of this label assignment is that it can not reflect the heterogeneous similarity of the query crop to crops from other images, but regarding them as equally negative. To address this issue, inspired by consistency regularization in semi-supervised learning, we propose Consistent Contrast (CO2), which introduces a consistency term into unsupervised contrastive learning framework. The consistency term takes the similarity of the query crop to crops from other images as unlabeled, and the corresponding similarity of a positive crop as a pseudo label. It then encourages consistency between these two similarities. Empirically, CO2 improves Momentum Contrast (MoCo) by 2.9% top-1 accuracy on ImageNet linear protocol, 3.8% and 1.1% top-5 accuracy on 1% and 10% labeled semi-supervised settings. It also transfers to image classification, object detection, and semantic segmentation on PASCAL VOC. This shows that CO2 learns better visual representations for downstream tasks.

NeurIPS Conference 2021 Conference Paper

Glance-and-Gaze Vision Transformer

• Qihang Yu
• Yingda Xia
• Yutong Bai
• Yongyi Lu
• Alan L. Yuille
• Wei Shen

Recently, there emerges a series of vision Transformers, which show superior performance with a more compact model size than conventional convolutional neural networks, thanks to the strong ability of Transformers to model long-range dependencies. However, the advantages of vision Transformers also come with a price: Self-attention, the core part of Transformer, has a quadratic complexity to the input sequence length. This leads to a dramatic increase of computation and memory cost with the increase of sequence length, thus introducing difficulties when applying Transformers to the vision tasks that require dense predictions based on high-resolution feature maps. In this paper, we propose a new vision Transformer, named Glance-and-Gaze Transformer (GG-Transformer), to address the aforementioned issues. It is motivated by the Glance and Gaze behavior of human beings when recognizing objects in natural scenes, with the ability to efficiently model both long-range dependencies and local context. In GG-Transformer, the Glance and Gaze behavior is realized by two parallel branches: The Glance branch is achieved by performing self-attention on the adaptively-dilated partitions of the input, which leads to a linear complexity while still enjoying a global receptive field; The Gaze branch is implemented by a simple depth-wise convolutional layer, which compensates local image context to the features obtained by the Glance mechanism. We empirically demonstrate our method achieves consistently superior performance over previous state-of-the-art Transformers on various vision tasks and benchmarks.

ICLR Conference 2021 Conference Paper

NeMo: Neural Mesh Models of Contrastive Features for Robust 3D Pose Estimation

• Angtian Wang
• Adam Kortylewski
• Alan L. Yuille

3D pose estimation is a challenging but important task in computer vision. In this work, we show that standard deep learning approaches to 3D pose estimation are not robust to partial occlusion. Inspired by the robustness of generative vision models to partial occlusion, we propose to integrate deep neural networks with 3D generative representations of objects into a unified neural architecture that we term NeMo. In particular, NeMo learns a generative model of neural feature activations at each vertex on a dense 3D mesh. Using differentiable rendering we estimate the 3D object pose by minimizing the reconstruction error between NeMo and the feature representation of the target image. To avoid local optima in the reconstruction loss, we train the feature extractor to maximize the distance between the individual feature representations on the mesh using contrastive learning. Our extensive experiments on PASCAL3D+, occluded-PASCAL3D+ and ObjectNet3D show that NeMo is much more robust to partial occlusion compared to standard deep networks, while retaining competitive performance on non-occluded data. Interestingly, our experiments also show that NeMo performs reasonably well even when the mesh representation only crudely approximates the true object geometry with a cuboid, hence revealing that the detailed 3D geometry is not needed for accurate 3D pose estimation.

NeurIPS Conference 2021 Conference Paper

Neural View Synthesis and Matching for Semi-Supervised Few-Shot Learning of 3D Pose

• Angtian Wang
• Shenxiao Mei
• Alan L. Yuille
• Adam Kortylewski

We study the problem of learning to estimate the 3D object pose from a few labelled examples and a collection of unlabelled data. Our main contribution is a learning framework, neural view synthesis and matching, that can transfer the 3D pose annotation from the labelled to unlabelled images reliably, despite unseen 3D views and nuisance variations such as the object shape, texture, illumination or scene context. In our approach, objects are represented as 3D cuboid meshes composed of feature vectors at each mesh vertex. The model is initialized from a few labelled images and is subsequently used to synthesize feature representations of unseen 3D views. The synthesized views are matched with the feature representations of unlabelled images to generate pseudo-labels of the 3D pose. The pseudo-labelled data is, in turn, used to train the feature extractor such that the features at each mesh vertex are more invariant across varying 3D views of the object. Our model is trained in an EM-type manner alternating between increasing the 3D pose invariance of the feature extractor and annotating unlabelled data through neural view synthesis and matching. We demonstrate the effectiveness of the proposed semi-supervised learning framework for 3D pose estimation on the PASCAL3D+ and KITTI datasets. We find that our approach outperforms all baselines by a wide margin, particularly in an extreme few-shot setting where only 7 annotated images are given. Remarkably, we observe that our model also achieves an exceptional robustness in out-of-distribution scenarios that involve partial occlusion.

ICLR Conference 2021 Conference Paper

Shape-Texture Debiased Neural Network Training

• Yingwei Li 0002
• Qihang Yu
• Mingxing Tan
• Jieru Mei
• Peng Tang
• Wei Shen 0002
• Alan L. Yuille
• Cihang Xie

Shape and texture are two prominent and complementary cues for recognizing objects. Nonetheless, Convolutional Neural Networks are often biased towards either texture or shape, depending on the training dataset. Our ablation shows that such bias degenerates model performance. Motivated by this observation, we develop a simple algorithm for shape-texture debiased learning. To prevent models from exclusively attending on a single cue in representation learning, we augment training data with images with conflicting shape and texture information (eg, an image of chimpanzee shape but with lemon texture) and, most importantly, provide the corresponding supervisions from shape and texture simultaneously. Experiments show that our method successfully improves model performance on several image recognition benchmarks and adversarial robustness. For example, by training on ImageNet, it helps ResNet-152 achieve substantial improvements on ImageNet (+1.2%), ImageNet-A (+5.2%), ImageNet-C (+8.3%) and Stylized-ImageNet (+11.1%), and on defending against FGSM adversarial attacker on ImageNet (+14.4%). Our method also claims to be compatible with other advanced data augmentation strategies, eg, Mixup, and CutMix. The code is available here: https://github.com/LiYingwei/ShapeTextureDebiasedTraining.

ICLR Conference 2020 Conference Paper

AtomNAS: Fine-Grained End-to-End Neural Architecture Search

• Jieru Mei
• Yingwei Li 0002
• Xiaochen Lian
• Xiaojie Jin
• Linjie Yang
• Alan L. Yuille
• Jianchao Yang

Search space design is very critical to neural architecture search (NAS) algorithms. We propose a fine-grained search space comprised of atomic blocks, a minimal search unit that is much smaller than the ones used in recent NAS algorithms. This search space allows a mix of operations by composing different types of atomic blocks, while the search space in previous methods only allows homogeneous operations. Based on this search space, we propose a resource-aware architecture search framework which automatically assigns the computational resources (e.g., output channel numbers) for each operation by jointly considering the performance and the computational cost. In addition, to accelerate the search process, we propose a dynamic network shrinkage technique which prunes the atomic blocks with negligible influence on outputs on the fly. Instead of a search-and-retrain two-stage paradigm, our method simultaneously searches and trains the target architecture. Our method achieves state-of-the-art performance under several FLOPs configurations on ImageNet with a small searching cost. We open our entire codebase at: https://github.com/meijieru/AtomNAS.

NeurIPS Conference 2020 Conference Paper

Every View Counts: Cross-View Consistency in 3D Object Detection with Hybrid-Cylindrical-Spherical Voxelization

• Qi Chen
• Lin Sun
• Ernest Cheung
• Alan L. Yuille

Recent voxel-based 3D object detectors for autonomous vehicles learn point cloud representations either from bird eye view (BEV) or range view (RV, a. k. a. the perspective view). However, each view has its own strengths and weaknesses. In this paper, we present a novel framework to unify and leverage the benefits from both BEV and RV. The widely-used cuboid-shaped voxels in Cartesian coordinate system only benefit learning BEV feature map. Therefore, to enable learning both BEV and RV feature maps, we introduce Hybrid-Cylindrical-Spherical voxelization. Our findings show that simply adding detection on another view as auxiliary supervision will lead to poor performance. We proposed a pair of cross-view transformers to transform the feature maps into the other view and introduce cross-view consistency loss on them. Comprehensive experiments on the challenging NuScenes Dataset validate the effectiveness of our proposed method by virtue of joint optimization and complementary information on both views. Remarkably, our approach achieved mAP of 55. 8%, outperforming all published approaches by at least 3% in overall performance and up to 16. 5% in safety-crucial categories like cyclist.

ICLR Conference 2020 Conference Paper

Intriguing Properties of Adversarial Training at Scale

• Cihang Xie
• Alan L. Yuille

Adversarial training is one of the main defenses against adversarial attacks. In this paper, we provide the first rigorous study on diagnosing elements of large-scale adversarial training on ImageNet, which reveals two intriguing properties. First, we study the role of normalization. Batch normalization (BN) is a crucial element for achieving state-of-the-art performance on many vision tasks, but we show it may prevent networks from obtaining strong robustness in adversarial training. One unexpected observation is that, for models trained with BN, simply removing clean images from training data largely boosts adversarial robustness, i.e., 18.3%. We relate this phenomenon to the hypothesis that clean images and adversarial images are drawn from two different domains. This two-domain hypothesis may explain the issue of BN when training with a mixture of clean and adversarial images, as estimating normalization statistics of this mixture distribution is challenging. Guided by this two-domain hypothesis, we show disentangling the mixture distribution for normalization, i.e., applying separate BNs to clean and adversarial images for statistics estimation, achieves much stronger robustness. Additionally, we find that enforcing BNs to behave consistently at training and testing can further enhance robustness. Second, we study the role of network capacity. We find our so-called "deep" networks are still shallow for the task of adversarial learning. Unlike traditional classification tasks where accuracy is only marginally improved by adding more layers to "deep" networks (e.g., ResNet-152), adversarial training exhibits a much stronger demand on deeper networks to achieve higher adversarial robustness. This robustness improvement can be observed substantially and consistently even by pushing the network capacity to an unprecedented scale, i.e., ResNet-638.

IROS Conference 2020 Conference Paper

Probabilistic Multi-modal Trajectory Prediction with Lane Attention for Autonomous Vehicles

• Chenxu Luo
• Lin Sun 0004
• Dariush Dabiri
• Alan L. Yuille

Trajectory prediction is crucial for autonomous vehicles. The planning system not only needs to know the current state of the surrounding objects but also their possible states in the future. As for vehicles, their trajectories are significantly influenced by the lane geometry and how to effectively use the lane information is of active interest. Most of the existing works use rasterized maps to explore road information, which does not distinguish different lanes. In this paper, we propose a novel instance-aware representation for lane representation. By integrating the lane features and trajectory features, a goal-oriented lane attention module is proposed to predict the future locations of the vehicle. We show that the proposed lane representation together with the lane attention module can be integrated into the widely used encoder-decoder framework to generate diverse predictions. Most importantly, each generated trajectory is associated with a probability to handle the uncertainty. Our method does not suffer from collapsing to one behavior modal and can cover diverse possibilities. Extensive experiments and ablation studies on the benchmark datasets corroborate the effectiveness of our proposed method. Notably, our proposed method ranks third place in the Argoverse motion forecasting competition at NeurIPS 2019 1.

AAAI Conference 2019 Conference Paper

Learning Basis Representation to Refine 3D Human Pose Estimations

• Chunyu Wang
• Haibo Qiu
• Alan L. Yuille
• Wenjun Zeng

Estimating 3D human poses from 2D joint positions is an illposed problem, and is further complicated by the fact that the estimated 2D joints usually have errors to which most of the 3D pose estimators are sensitive. In this work, we present an approach to refine inaccurate 3D pose estimations. The core idea of the approach is to learn a number of bases to obtain tight approximations of the low-dimensional pose manifold where a 3D pose is represented by a convex combination of the bases. The representation requires that globally the refined poses are close to the pose manifold thus avoiding generating illegitimate poses. Second, the designed bases also have the property to guarantee that the distances among the body joints of a pose are within reasonable ranges. Experiments on benchmark datasets show that our approach obtains more legitimate poses over the baselines. In particular, the limb lengths are closer to the ground truth.

AAAI Conference 2019 Conference Paper

Training Deep Neural Networks in Generations: A More Tolerant Teacher Educates Better Students

• Chenglin Yang
• Lingxi Xie
• Siyuan Qiao
• Alan L. Yuille

We focus on the problem of training a deep neural network in generations. The flowchart is that, in order to optimize the target network (student), another network (teacher) with the same architecture is first trained, and used to provide part of supervision signals in the next stage. While this strategy leads to a higher accuracy, many aspects (e. g. , why teacher-student optimization helps) still need further explorations. This paper studies this problem from a perspective of controlling the strictness in training the teacher network. Existing approaches mostly used a hard distribution (e. g. , one-hot vectors) in training, leading to a strict teacher which itself has a high accuracy, but we argue that the teacher needs to be more tolerant, although this often implies a lower accuracy. The implementation is very easy, with merely an extra loss term added to the teacher network, facilitating a few secondary classes to emerge and complement to the primary class. Consequently, the teacher provides a milder supervision signal (a less peaked distribution), and makes it possible for the student to learn from inter-class similarity and potentially lower the risk of over-fitting. Experiments are performed on standard image classification tasks (CIFAR100 and ILSVRC2012). Although the teacher network behaves less powerful, the students show a persistent ability growth and eventually achieve higher classification accuracies than other competitors. Model ensemble and transfer feature extraction also verify the effectiveness of our approach.

ICML Conference 2018 Conference Paper

Gradually Updated Neural Networks for Large-Scale Image Recognition

• Siyuan Qiao
• Zhishuai Zhang
• Wei Shen 0002
• Bo Wang 0044
• Alan L. Yuille

Depth is one of the keys that make neural networks succeed in the task of large-scale image recognition. The state-of-the-art network architectures usually increase the depths by cascading convolutional layers or building blocks. In this paper, we present an alternative method to increase the depth. Our method is by introducing computation orderings to the channels within convolutional layers or blocks, based on which we gradually compute the outputs in a channel-wise manner. The added orderings not only increase the depths and the learning capacities of the networks without any additional computation costs, but also eliminate the overlap singularities so that the networks are able to converge faster and perform better. Experiments show that the networks based on our method achieve the state-of-the-art performances on CIFAR and ImageNet datasets.

ICLR Conference 2015 Conference Paper

Deep Captioning with Multimodal Recurrent Neural Networks (m-RNN)

• Junhua Mao
• Wei Xu 0017
• Yi Yang 0007
• Jiang Wang 0001
• Alan L. Yuille

In this paper, we present a multimodal Recurrent Neural Network (m-RNN) model for generating novel image captions. It directly models the probability distribution of generating a word given previous words and an image. Image captions are generated by sampling from this distribution. The model consists of two sub-networks: a deep recurrent neural network for sentences and a deep convolutional network for images. These two sub-networks interact with each other in a multimodal layer to form the whole m-RNN model. The effectiveness of our model is validated on four benchmark datasets (IAPR TC-12, Flickr 8K, Flickr 30K and MS COCO). Our model outperforms the state-of-the-art methods. In addition, we apply the m-RNN model to retrieval tasks for retrieving images or sentences, and achieves significant performance improvement over the state-of-the-art methods which directly optimize the ranking objective function for retrieval. The project page of this work is: www.stat.ucla.edu/~junhua.mao/m-RNN.html .

ICML Conference 2015 Conference Paper

Learning Deep Structured Models

• Liang-Chieh Chen
• Alexander G. Schwing
• Alan L. Yuille
• Raquel Urtasun

Many problems in real-world applications involve predicting several random variables that are statistically related. Markov random fields (MRFs) are a great mathematical tool to encode such dependencies. The goal of this paper is to combine MRFs with deep learning to estimate complex representations while taking into account the dependencies between the output random variables. Towards this goal, we propose a training algorithm that is able to learn structured models jointly with deep features that form the MRF potentials. Our approach is efficient as it blends learning and inference and makes use of GPU acceleration. We demonstrate the effectiveness of our algorithm in the tasks of predicting words from noisy images, as well as tagging of Flickr photographs. We show that joint learning of the deep features and the MRF parameters results in significant performance gains.

YNIMG Journal 2014 Journal Article

Non-negative matrix factorization of multimodal MRI, fMRI and phenotypic data reveals differential changes in default mode subnetworks in ADHD

• Ariana Anderson
• Pamela K. Douglas
• Wesley T. Kerr
• Virginia S. Haynes
• Alan L. Yuille
• Jianwen Xie
• Ying Nian Wu
• Jesse A. Brown

In the multimodal neuroimaging framework, data on a single subject are collected from inherently different sources such as functional MRI, structural MRI, behavioral and/or phenotypic information. The information each source provides is not independent; a subset of features from each modality maps to one or more common latent dimensions, which can be interpreted using generative models. These latent dimensions, or “topics, ” provide a sparse summary of the generative process behind the features for each individual. Topic modeling, an unsupervised generative model, has been used to map seemingly disparate features to a common domain. We use Non-Negative Matrix Factorization (NMF) to infer the latent structure of multimodal ADHD data containing fMRI, MRI, phenotypic and behavioral measurements. We compare four different NMF algorithms and find that the sparsest decomposition is also the most differentiating between ADHD and healthy patients. We identify dimensions that map to interpretable, recognizable dimensions such as motion, default mode network activity, and other such features of the input data. For example, structural and functional graph theory features related to default mode subnetworks clustered with the ADHD-Inattentive diagnosis. Structural measurements of the default mode network (DMN) regions such as the posterior cingulate, precuneus, and parahippocampal regions were all related to the ADHD-Inattentive diagnosis. Ventral DMN subnetworks may have more functional connections in ADHD-I, while dorsal DMN may have less. ADHD topics are dependent upon diagnostic site, suggesting diagnostic differences across geographic locations. We assess our findings in light of the ADHD-200 classification competition, and contrast our unsupervised, nominated topics with previously published supervised learning methods. Finally, we demonstrate the validity of these latent variables as biomarkers by using them for classification of ADHD in 730 patients. Cumulatively, this manuscript addresses how multimodal data in ADHD can be interpreted by latent dimensions.

ICLR Conference 2013 Conference Paper

Complexity of Representation and Inference in Compositional Models with Part Sharing

• Alan L. Yuille
• Roozbeh Mottaghi

This paper describes serial and parallel compositional models of multiple objects with part sharing. Objects are built by part-subpart compositions and expressed in terms of a hierarchical dictionary of object parts. These parts are represented on lattices of decreasing sizes which yield an executive summary description. We describe inference and learning algorithms for these models. We analyze the complexity of this model in terms of computation time (for serial computers) and numbers of nodes (e.g., 'neurons') for parallel computers. In particular, we compute the complexity gains by part sharing and its dependence on how the dictionary scales with the level of the hierarchy. We explore three regimes of scaling behavior where the dictionary size (i) increases exponentially with the level, (ii) is determined by an unsupervised compositional learning algorithm applied to real data, (iii) decreases exponentially with scale. This analysis shows that in some regimes the use of shared parts enables algorithms which can perform inference in time linear in the number of levels for an exponential number of objects. In other regimes part sharing has little advantage for serial computers but can give linear processing on parallel computers.

ICML Conference 2009 Conference Paper

Compositional noisy-logical learning

• Alan L. Yuille
• Songfeng Zheng

UAI Conference 2005 Conference Paper

The DLR Hierarchy of Approximate Inference

• Michal Rosen-Zvi
• Michael I. Jordan
• Alan L. Yuille

We propose a hierarchy for approximate inference based on the Dobrushin, Lanford, Ruelle (DLR) equations. This hierarchy includes existing algorithms, such as belief propagation, and also motivates novel algorithms such as factorized neighbors (FN) algorithms and variants of mean field (MF) algorithms. In particular, we show that extrema of the Bethe free energy correspond to approximate solutions of the DLR equations. In addition, we demonstrate a close connection between these approximate algorithms and Gibbs sampling. Finally, we compare and contrast various of the algorithms in the DLR hierarchy on spin-glass problems. The experiments show that algorithms higher up in the hierarchy give more accurate results when they converge but tend to be less stable.

AIJ Journal 1995 Journal Article

Multilevel enhancement and detection of stereo disparity surfaces

• Yibing Yang
• Alan L. Yuille

The problem of stereo vision has been of increasing interest to the computer vision community over the past decade. This paper presents a new computational framework for matching a pair of stereo images arising from viewing the same object from two different positions. In contrast to previous work, this approach formulates the matching problem as detection of a “bright”, coherent disparity surface in a 3D image called the spatio-disparity space (SDS) image. The SDS images represents the goodness of each and every possible match. A nonlinear filter is proposed for enhancing the disparity surface in the SDS image and for suppressing the noise. This filter is used to construct a hyperpyramid representation of the SDS image. Then the disparity surface is detected using a coarse-to-fine control structure. The proposed method is robust to photometric and geometric distortions in the stereo images, and has a number of computational advantages. It produces good results for complex scenes.

ICRA Conference 1995 Conference Paper

Parameterized Surface Fitting via MAP Estimation for Binocular Stereo

• Michael J. Weisman
• Alan L. Yuille
• James J. Clark

We present a novel method for reconstructing three dimensional surfaces from stereo intensity data. We employ a set of competing surface hypotheses based on parameterized models. We use maximum a posteriori (MAP) estimation and demonstrate a connection to the Hough transform. Experimental results are given showing the effectiveness of the algorithm.

ECAI Conference 1985 Conference Paper

The Smoothest Velocity Field Token Matching Schemes

• Alan L. Yuille