Ming C. Lin

Nataniel Ruiz, Miriam Bellver, Timo Bolkart et al. · amazon-science

We present a Body Measurement network (BMnet) for estimating 3D anthropomorphic measurements of the human body shape from silhouette images. Training of BMnet is performed on data from real human subjects, and augmented with a novel adversarial body simulator (ABS) that finds and synthesizes challenging body shapes. ABS is based on the skinned multiperson linear (SMPL) body model, and aims to maximize BMnet measurement prediction error with respect to latent SMPL shape parameters. ABS is fully differentiable with respect to these parameters, and trained end-to-end via backpropagation with BMnet in the loop. Experiments show that ABS effectively discovers adversarial examples, such as bodies with extreme body mass indices (BMI), consistent with the rarity of extreme-BMI bodies in BMnet's training set. Thus ABS is able to reveal gaps in training data and potential failures in predicting under-represented body shapes. Results show that training BMnet with ABS improves measurement prediction accuracy on real bodies by up to 10%, when compared to no augmentation or random body shape sampling. Furthermore, our method significantly outperforms SOTA measurement estimation methods by as much as 3x. Finally, we release BodyM, the first challenging, large-scale dataset of photo silhouettes and body measurements of real human subjects, to further promote research in this area. Project website: https://adversarialbodysim.github.io

Divya Kothandaraman, Tianyi Zhou, Ming Lin et al.

We present HawkI, for synthesizing aerial-view images from text and an exemplar image, without any additional multi-view or 3D information for finetuning or at inference. HawkI uses techniques from classical computer vision and information theory. It seamlessly blends the visual features from the input image within a pretrained text-to-2Dimage stable diffusion model with a test-time optimization process for a careful bias-variance trade-off, which uses an Inverse Perspective Mapping (IPM) homography transformation to provide subtle cues for aerialview synthesis. At inference, HawkI employs a unique mutual information guidance formulation to steer the generated image towards faithfully replicating the semantic details of the input-image, while maintaining a realistic aerial perspective. Mutual information guidance maximizes the semantic consistency between the generated image and the input image, without enforcing pixel-level correspondence between vastly different viewpoints. Through extensive qualitative and quantitative comparisons against text + exemplar-image based methods and 3D/ multi-view based novel-view synthesis methods on proposed synthetic and real datasets, we demonstrate that our method achieves a significantly better bias-variance trade-off towards generating high fidelity aerial-view images.Code and data is available at https://github.com/divyakraman/HawkI2024.

Yi-Ling Qiao, Alexander Gao, Ming C. Lin

We present a method for learning 3D geometry and physics parameters of a dynamic scene from only a monocular RGB video input. To decouple the learning of underlying scene geometry from dynamic motion, we represent the scene as a time-invariant signed distance function (SDF) which serves as a reference frame, along with a time-conditioned deformation field. We further bridge this neural geometry representation with a differentiable physics simulator by designing a two-way conversion between the neural field and its corresponding hexahedral mesh, enabling us to estimate physics parameters from the source video by minimizing a cycle consistency loss. Our method also allows a user to interactively edit 3D objects from the source video by modifying the recovered hexahedral mesh, and propagating the operation back to the neural field representation. Experiments show that our method achieves superior mesh and video reconstruction of dynamic scenes compared to competing Neural Field approaches, and we provide extensive examples which demonstrate its ability to extract useful 3D representations from videos captured with consumer-grade cameras.

Xijun Wang, Junbang Liang, Chun-Kai Wang et al.

In this work, we propose an efficient Video-Language Alignment (ViLA) network. Our ViLA model addresses both efficient frame sampling and effective cross-modal alignment in a unified way. In our ViLA network, we design a new learnable text-guided Frame-Prompter together with a new cross-modal distillation (QFormer-Distiller) module. Pre-trained large image-language models have shown promising results on problems such as visual question answering (VQA). However, how to efficiently and effectively sample video frames when adapting pre-trained large image-language model to video-language alignment is still the major challenge. Compared with prior work, our ViLA model demonstrates the capability of selecting key frames with critical contents, thus improving the video-language alignment accuracy while reducing the inference latency +3.3% on NExT-QA Temporal with 3.0X speed up). Overall, our ViLA network outperforms the state-of-the-art methods on the video question-answering benchmarks: +4.6% on STAR Interaction, +2.2% on STAR average with 3.0X speed up, ours 2-frames out-perform SeViLA 4-frames on the VLEP dataset with 4.2X speed-up. The code will be available at https://github.com/xijun-cs/ViLA.

Sanghyun Son, Laura Yu Zheng, Ryan Sullivan et al.

We introduce a novel policy learning method that integrates analytical gradients from differentiable environments with the Proximal Policy Optimization (PPO) algorithm. To incorporate analytical gradients into the PPO framework, we introduce the concept of an α-policy that stands as a locally superior policy. By adaptively modifying the α value, we can effectively manage the influence of analytical policy gradients during learning. To this end, we suggest metrics for assessing the variance and bias of analytical gradients, reducing dependence on these gradients when high variance or bias is detected. Our proposed approach outperforms baseline algorithms in various scenarios, such as function optimization, physics simulations, and traffic control environments. Our code can be found online: https://github.com/SonSang/gippo.

Xijun Wang, Anqi Liang, Junbang Liang et al.

Scene-aware Complementary Item Retrieval (CIR) is a challenging task which requires to generate a set of compatible items across domains. Due to the subjectivity, it is difficult to set up a rigorous standard for both data collection and learning objectives. To address this challenging task, we propose a visual compatibility concept, composed of similarity (resembling in color, geometry, texture, and etc.) and complementarity (different items like table vs chair completing a group). Based on this notion, we propose a compatibility learning framework, a category-aware Flexible Bidirectional Transformer (FBT), for visual "scene-based set compatibility reasoning" with the cross-domain visual similarity input and auto-regressive complementary item generation. We introduce a "Flexible Bidirectional Transformer (FBT)" consisting of an encoder with flexible masking, a category prediction arm, and an auto-regressive visual embedding prediction arm. And the inputs for FBT are cross-domain visual similarity invariant embeddings, making this framework quite generalizable. Furthermore, our proposed FBT model learns the inter-object compatibility from a large set of scene images in a self-supervised way. Compared with the SOTA methods, this approach achieves up to 5.3% and 9.6% in FITB score and 22.3% and 31.8% SFID improvement on fashion and furniture, respectively.

Divya Kothandaraman, Ming Lin, Dinesh Manocha

We introduce a novel approach for concept blending in pretrained text-to-image diffusion models, aiming to generate images at the intersection of multiple text prompts. At each time step during diffusion denoising, our algorithm forecasts predictions w.r.t. the generated image and makes informed text conditioning decisions. Central to our method is the unique analogy between diffusion models, which are rooted in non-equilibrium thermodynamics, and the Black-Scholes model for financial option pricing. By drawing parallels between key variables in both domains, we derive a robust algorithm for concept blending that capitalizes on the Markovian dynamics of the Black-Scholes framework. Our text-based concept blending algorithm is data-efficient, meaning it does not need additional training. Furthermore, it operates without human intervention or hyperparameter tuning. We highlight the benefits of our approach by comparing it qualitatively and quantitatively to other text based concept blending techniques, including linear interpolation, alternating prompts, step-wise prompt switching, and CLIP-guided prompt selection across various scenarios such as single object per text prompt, multiple objects per text prompt and objects against backgrounds. Our work shows that financially inspired techniques can enhance text-to-image concept blending in generative AI, paving the way for broader innovation. Code is available at https://github.com/divyakraman/BlackScholesDiffusion2024.

Anh N. Nhu, Sanghyun Son, Ming Lin

In this work, we introduce the Time-Aware World Model (TAWM), a model-based approach that explicitly incorporates temporal dynamics. By conditioning on the time-step size, Δt, and training over a diverse range of Δt values -- rather than sampling at a fixed time-step -- TAWM learns both high- and low-frequency task dynamics across diverse control problems. Grounded in the information-theoretic insight that the optimal sampling rate depends on a system's underlying dynamics, this time-aware formulation improves both performance and data efficiency. Empirical evaluations show that TAWM consistently outperforms conventional models across varying observation rates in a variety of control tasks, using the same number of training samples and iterations. Our code can be found online at: github.com/anh-nn01/Time-Aware-World-Model.

Sanghyun Son, Laura Zheng, Brian Clipp et al.

We present a parallelized differentiable traffic simulator based on the Intelligent Driver Model (IDM), a car-following framework that incorporates driver behavior as key variables. Our vehicle simulator efficiently models vehicle motion, generating trajectories that can be supervised to fit real-world data. By leveraging its differentiable nature, IDM parameters are optimized using gradient-based methods. With the capability to simulate up to 2 million vehicles in real time, the system is scalable for large-scale trajectory optimization. We show that we can use the simulator to filter noise in the input trajectories (trajectory filtering), reconstruct dense trajectories from sparse ones (trajectory reconstruction), and predict future trajectories (trajectory prediction), with all generated trajectories adhering to physical laws. We validate our simulator and algorithm on several datasets including NGSIM and Waymo Open Dataset. The code is publicly available at: https://github.com/SonSang/diffidm.

Sanghyun Son, Matheus Gadelha, Yang Zhou et al.

We present a differentiable representation, DMesh, for general 3D triangular meshes. DMesh considers both the geometry and connectivity information of a mesh. In our design, we first get a set of convex tetrahedra that compactly tessellates the domain based on Weighted Delaunay Triangulation (WDT), and select triangular faces on the tetrahedra to define the final mesh. We formulate probability of faces to exist on the actual surface in a differentiable manner based on the WDT. This enables DMesh to represent meshes of various topology in a differentiable way, and allows us to reconstruct the mesh under various observations, such as point cloud and multi-view images using gradient-based optimization. The source code and full paper is available at: https://sonsang.github.io/dmesh-project.

Sanghyun Son, Matheus Gadelha, Yang Zhou et al.

Recent probabilistic methods for 3D triangular meshes capture diverse shapes by differentiable mesh connectivity, but face high computational costs with increased shape details. We introduce a new differentiable mesh processing method that addresses this challenge and efficiently handles meshes with intricate structures. Our method reduces time complexity from O(N) to O(log N) and requires significantly less memory than previous approaches. Building on this innovation, we present a reconstruction algorithm capable of generating complex 2D and 3D shapes from point clouds or multi-view images. Visit our project page (https://sonsang.github.io/dmesh2-project) for source code and supplementary material.

Gang Li, Yan Chen, Ming Lin et al.

Recent large reasoning models (LRMs) driven by reinforcement learning algorithms (e.g., GRPO) have achieved remarkable performance on challenging reasoning tasks. However, these models suffer from overthinking, generating unnecessarily long and redundant reasoning even for simple questions, which substantially increases computational cost and response latency. While existing methods incorporate length rewards to GRPO to promote concise reasoning, they incur significant performance degradation. We identify the root cause: when rewards for correct but long rollouts are penalized, GRPO's group-relative advantage function can assign them negative advantages, actively discouraging valid reasoning. To overcome this, we propose Decoupled Reward Policy Optimization (DRPO), a novel framework that decouples the length-based learning signal of correct rollouts from incorrect ones. DRPO ensures that reward signals for correct rollouts are normalized solely within the positive group, shielding them from interference by negative samples. The DRPO's objective is grounded in integrating an optimized positive data distribution, which maximizes length-based rewards under a KL regularization, into a discriminative objective. We derive a closed-form solution for this distribution, enabling efficient computation of the objective and its gradients using only on-policy data and importance weighting. Of independent interest, this formulation is general and can incorporate other preference rewards of positive data beyond length. Experiments on mathematical reasoning tasks demonstrate DRPO's significant superiority over six efficient reasoning baselines. Notably, with a 1.5B model, our method achieves 77\% length reduction with only 1.1\% performance loss on simple questions like GSM8k dataset, while the follow-up baseline sacrifices 4.3\% for 68\% length reduction.

Shreelekha Revankar, Shijia Liao, Yu Shen et al.

The accuracy of 3D Human Pose and Shape reconstruction (HPS) from an image is progressively improving. Yet, no known method is robust across all image distortion. To address issues due to variations of camera poses, we introduce SHARE, a novel fine-tuning method that utilizes adversarial data augmentation to enhance the robustness of existing HPS techniques. We perform a comprehensive analysis on the impact of camera poses on HPS reconstruction outcomes. We first generated large-scale image datasets captured systematically from diverse camera perspectives. We then established a mapping between camera poses and reconstruction errors as a continuous function that characterizes the relationship between camera poses and HPS quality. Leveraging this representation, we introduce RoME (Regions of Maximal Error), a novel sampling technique for our adversarial fine-tuning method. The SHARE framework is generalizable across various single-view HPS methods and we demonstrate its performance on HMR, SPIN, PARE, CLIFF and ExPose. Our results illustrate a reduction in mean joint errors across single-view HPS techniques, for images captured from multiple camera positions without compromising their baseline performance. In many challenging cases, our method surpasses the performance of existing models, highlighting its practical significance for diverse real-world applications.

Justin Wilson, Sunyeong Park, Seunghye J. Wilson et al.

Face aging techniques have used generative adversarial networks (GANs) and style transfer learning to transform one's appearance to look younger/older. Identity is maintained by conditioning these generative networks on a learned vector representation of the source content. In this work, we apply a similar approach to age a speaker's voice, referred to as voice aging. We first analyze the classification of a speaker's age by training a convolutional neural network (CNN) on the speaker's voice and face data from Common Voice and VoxCeleb datasets. We generate aged voices from style transfer to transform an input spectrogram to various ages and demonstrate our method on a mobile app.

Justin Wilson, Nicholas Rewkowski, Ming C. Lin et al.

Reflective and textureless surfaces such as windows, mirrors, and walls can be a challenge for object and scene reconstruction. These surfaces are often poorly reconstructed and filled with depth discontinuities and holes, making it difficult to cohesively reconstruct scenes that contain these planar discontinuities. We propose Echoreconstruction, an audio-visual method that uses the reflections of sound to aid in geometry and audio reconstruction for virtual conferencing, teleimmersion, and other AR/VR experience. The mobile phone prototype emits pulsed audio, while recording video for RGB-based 3D reconstruction and audio-visual classification. Reflected sound and images from the video are input into our audio (EchoCNN-A) and audio-visual (EchoCNN-AV) convolutional neural networks for surface and sound source detection, depth estimation, and material classification. The inferences from these classifications enhance scene 3D reconstructions containing open spaces and reflective surfaces by depth filtering, inpainting, and placement of unmixed sound sources in the scene. Our prototype, VR demo, and experimental results from real-world and virtual scenes with challenging surfaces and sound indicate high success rates on classification of material, depth estimation, and closed/open surfaces, leading to considerable visual and audio improvement in 3D scenes (see Figure 1).

Justin Wilson, Ming C. Lin

3D object reconstructions of transparent and concave structured objects, with inferred material properties, remains an open research problem for robot navigation in unstructured environments. In this paper, we propose a multimodal single- and multi-frame neural network for 3D reconstructions using audio-visual inputs. Our trained reconstruction LSTM autoencoder 3D-MOV accepts multiple inputs to account for a variety of surface types and views. Our neural network produces high-quality 3D reconstructions using voxel representation. Based on Intersection-over-Union (IoU), we evaluate against other baseline methods using synthetic audio-visual datasets ShapeNet and Sound20K with impact sounds and bounding box annotations. To the best of our knowledge, our single- and multi-frame model is the first audio-visual reconstruction neural network for 3D geometry and material representation.

Shivam Akhauri, Laura Zheng, Tom Goldstein et al.

Practical learning-based autonomous driving models must be capable of generalizing learned behaviors from simulated to real domains, and from training data to unseen domains with unusual image properties. In this paper, we investigate transfer learning methods that achieve robustness to domain shifts by taking advantage of the invariance of spatio-temporal features across domains. In this paper, we propose a transfer learning method to improve generalization across domains via transfer of spatio-temporal features and salient data augmentation. Our model uses a CNN-LSTM network with Inception modules for image feature extraction. Our method runs in two phases: Phase 1 involves training on source domain data, while Phase 2 performs training on target domain data that has been supplemented by feature maps generated using the Phase 1 model. Our model significantly improves performance in unseen test cases for both simulation-to-simulation transfer as well as simulation-to-real transfer by up to +37.3\% in test accuracy and up to +40.8\% in steering angle prediction, compared to other SOTA methods across multiple datasets.

Yu Shen, Laura Zheng, Manli Shu et al.

For safety of autonomous driving, vehicles need to be able to drive under various lighting, weather, and visibility conditions in different environments. These external and environmental factors, along with internal factors associated with sensors, can pose significant challenges to perceptual data processing, hence affecting the decision-making and control of the vehicle. In this work, we address this critical issue by introducing a framework for analyzing robustness of the learning algorithm w.r.t varying quality in the image input for autonomous driving. Using the results of sensitivity analysis, we further propose an algorithm to improve the overall performance of the task of "learning to steer". The results show that our approach is able to enhance the learning outcomes up to 48%. A comparative study drawn between our approach and other related techniques, such as data augmentation and adversarial training, confirms the effectiveness of our algorithm as a way to improve the robustness and generalization of neural network training for autonomous driving.

Shivang Patel, Senthil Hariharan, Pranav Dhulipala et al.

We study multi-agent coverage algorithms for autonomous monitoring and patrol in urban environments. We consider scenarios in which a team of flying agents uses downward facing cameras (or similar sensors) to observe the environment outside of buildings at street-level. Buildings are considered obstacles that impede movement, and cameras are assumed to be ineffective above a maximum altitude. We study multi-agent urban coverage problems related to this scenario, including: (1) static multi-agent urban coverage, in which agents are expected to observe the environment from static locations, and (2) dynamic multi-agent urban coverage where agents move continuously through the environment. We experimentally evaluate six different multi-agent coverage methods, including: three types of ergodic coverage (that avoid buildings in different ways), lawn-mower sweep, voronoi region based control, and a naive grid method. We evaluate all algorithms with respect to four performance metrics (percent coverage, revist count, revist time, and the integral of area viewed over time), across four types of urban environments [low density, high density] x [short buildings, tall buildings], and for team sizes ranging from 2 to 25 agents. We believe this is the first extensive comparison of these methods in an urban setting. Our results highlight how the relative performance of static and dynamic methods changes based on the ratio of team size to search area, as well the relative effects that different characteristics of urban environments (tall, short, dense, sparse, mixed) have on each algorithm.

Shivam Akhauri, Laura Zheng, Ming Lin

Simulation data can be utilized to extend real-world driving data in order to cover edge cases, such as vehicle accidents. The importance of handling edge cases can be observed in the high societal costs in handling car accidents, as well as potential dangers to human drivers. In order to cover a wide and diverse range of all edge cases, we systemically parameterize and simulate the most common accident scenarios. By applying this data to autonomous driving models, we show that transfer learning on simulated data sets provide better generalization and collision avoidance, as compared to random initialization methods. Our results illustrate that information from a model trained on simulated data can be inferred to a model trained on real-world data, indicating the potential influence of simulation data in real world models and advancements in handling of anomalous driving scenarios.

Yi-Ling Qiao, Junbang Liang, Vladlen Koltun et al.

Differentiable physics is a powerful approach to learning and control problems that involve physical objects and environments. While notable progress has been made, the capabilities of differentiable physics solvers remain limited. We develop a scalable framework for differentiable physics that can support a large number of objects and their interactions. To accommodate objects with arbitrary geometry and topology, we adopt meshes as our representation and leverage the sparsity of contacts for scalable differentiable collision handling. Collisions are resolved in localized regions to minimize the number of optimization variables even when the number of simulated objects is high. We further accelerate implicit differentiation of optimization with nonlinear constraints. Experiments demonstrate that the presented framework requires up to two orders of magnitude less memory and computation in comparison to recent particle-based methods. We further validate the approach on inverse problems and control scenarios, where it outperforms derivative-free and model-free baselines by at least an order of magnitude.

Junbang Liang, Ming C. Lin

We propose a scalable neural network framework to reconstruct the 3D mesh of a human body from multi-view images, in the subspace of the SMPL model. Use of multi-view images can significantly reduce the projection ambiguity of the problem, increasing the reconstruction accuracy of the 3D human body under clothing. Our experiments show that this method benefits from the synthetic dataset generated from our pipeline since it has good flexibility of variable control and can provide ground-truth for validation. Our method outperforms existing methods on real-world images, especially on shape estimations.

Weizi Li, David Wolinski, Ming C. Lin

Autonomous driving has gained significant advancements in recent years. However, obtaining a robust control policy for driving remains challenging as it requires training data from a variety of scenarios, including rare situations (e.g., accidents), an effective policy architecture, and an efficient learning mechanism. We propose ADAPS for producing robust control policies for autonomous vehicles. ADAPS consists of two simulation platforms in generating and analyzing accidents to automatically produce labeled training data, and a memory-enabled hierarchical control policy. Additionally, ADAPS offers a more efficient online learning mechanism that reduces the number of iterations required in learning compared to existing methods such as DAGGER. We present both theoretical and experimental results. The latter are produced in simulated environments, where qualitative and quantitative results are generated to demonstrate the benefits of ADAPS.

Senthil Hariharan Arul, Adarsh Jagan Sathyamoorthy, Shivang Patel et al.

In this paper, we address the problem of collision avoidance for a swarm of UAVs used for continuous surveillance of an urban environment. Our method, LSwarm, efficiently avoids collisions with static obstacles, dynamic obstacles and other agents in 3-D urban environments while considering coverage constraints. LSwarm computes collision avoiding velocities that (i) maximize the conformity of an agent to an optimal path given by a global coverage strategy and (ii) ensure sufficient resolution of the coverage data collected by each agent. Our algorithm is formulated based on ORCA (Optimal Reciprocal Collision Avoidance) and is scalable with respect to the size of the swarm. We evaluate the coverage performance of LSwarm in realistic simulations of a swarm of quadrotors in complex urban models. In practice, our approach can compute collision avoiding velocities for a swarm composed of tens to hundreds of agents in a few milliseconds on dense urban scenes consisting of tens of buildings.