Where do Large Vision-Language Models Look at when Answering Questions?

Large Vision-Language Models (LVLMs) have shown promising performance in vision-language understanding and reasoning tasks. However, their visual understanding behaviors remain underexplored. A fundamental question arises: to what extent do LVLMs rely on visual input, and which image regions contribute to their responses? It is non-trivial to interpret the free-form generation of LVLMs due to their complicated visual architecture (e.g., multiple encoders and multi-resolution) and variable-length outputs. In this paper, we extend existing heatmap visualization methods (e.g., iGOS++) to support LVLMs for open-ended visual question answering. We propose a method to select visually relevant tokens that reflect the relevance between generated answers and input image. Furthermore, we conduct a comprehensive analysis of state-of-the-art LVLMs on benchmarks designed to require visual information to answer. Our findings offer several insights into LVLM behavior, including the relationship between focus region and answer correctness, differences in visual attention across architectures, and the impact of LLM scale on visual understanding. The code and data are available at https://github.com/bytedance/LVLM_Interpretation.

PointRecon: Online Point-based 3D Reconstruction via Ray-based 2D-3D Matching

We propose a novel online, point-based 3D reconstruction method from posed monocular RGB videos. Our model maintains a global point cloud representation of the scene, continuously updating the features and 3D locations of points as new images are observed. It expands the point cloud with newly detected points while carefully removing redundancies. The point cloud updates and depth predictions for new points are achieved through a novel ray-based 2D-3D feature matching technique, which is robust against errors in previous point position predictions. In contrast to offline methods, our approach processes infinite-length sequences and provides real-time updates. Additionally, the point cloud imposes no pre-defined resolution or scene size constraints, and its unified global representation ensures view consistency across perspectives. Experiments on the ScanNet dataset show that our method achieves state-of-the-art quality among online MVS approaches. Project page: https://arthurhero.github.io/projects/pointrecon

Long-LRM: Long-sequence Large Reconstruction Model for Wide-coverage Gaussian Splats

We propose Long-LRM, a generalizable 3D Gaussian reconstruction model that is capable of reconstructing a large scene from a long sequence of input images. Specifically, our model can process 32 source images at 960x540 resolution within only 1.3 seconds on a single A100 80G GPU. Our architecture features a mixture of the recent Mamba2 blocks and the classical transformer blocks which allowed many more tokens to be processed than prior work, enhanced by efficient token merging and Gaussian pruning steps that balance between quality and efficiency. Unlike previous feed-forward models that are limited to processing 1~4 input images and can only reconstruct a small portion of a large scene, Long-LRM reconstructs the entire scene in a single feed-forward step. On large-scale scene datasets such as DL3DV-140 and Tanks and Temples, our method achieves performance comparable to optimization-based approaches while being two orders of magnitude more efficient. Project page: https://arthurhero.github.io/projects/llrm

Long-Term 3D Point Tracking By Cost Volume Fusion

Long-term point tracking is essential to understand non-rigid motion in the physical world better. Deep learning approaches have recently been incorporated into long-term point tracking, but most prior work predominantly functions in 2D. Although these methods benefit from the well-established backbones and matching frameworks, the motions they produce do not always make sense in the 3D physical world. In this paper, we propose the first deep learning framework for long-term point tracking in 3D that generalizes to new points and videos without requiring test-time fine-tuning. Our model contains a cost volume fusion module that effectively integrates multiple past appearances and motion information via a transformer architecture, significantly enhancing overall tracking performance. In terms of 3D tracking performance, our model significantly outperforms simple scene flow chaining and previous 2D point tracking methods, even if one uses ground truth depth and camera pose to backproject 2D point tracks in a synthetic scenario.

Point Cloud Models Improve Visual Robustness in Robotic Learners

Visual control policies can encounter significant performance degradation when visual conditions like lighting or camera position differ from those seen during training -- often exhibiting sharp declines in capability even for minor differences. In this work, we examine robustness to a suite of these types of visual changes for RGB-D and point cloud based visual control policies. To perform these experiments on both model-free and model-based reinforcement learners, we introduce a novel Point Cloud World Model (PCWM) and point cloud based control policies. Our experiments show that policies that explicitly encode point clouds are significantly more robust than their RGB-D counterparts. Further, we find our proposed PCWM significantly outperforms prior works in terms of sample efficiency during training. Taken together, these results suggest reasoning about the 3D scene through point clouds can improve performance, reduce learning time, and increase robustness for robotic learners. Project Webpage: https://pvskand.github.io/projects/PCWM

Object Dynamics Modeling with Hierarchical Point Cloud-based Representations

Modeling object dynamics with a neural network is an important problem with numerous applications. Most recent work has been based on graph neural networks. However, physics happens in 3D space, where geometric information potentially plays an important role in modeling physical phenomena. In this work, we propose a novel U-net architecture based on continuous point convolution which naturally embeds information from 3D coordinates and allows for multi-scale feature representations with established downsampling and upsampling procedures. Bottleneck layers in the downsampled point clouds lead to better long-range interaction modeling. Besides, the flexibility of point convolutions allows our approach to generalize to sparsely sampled points from mesh vertices and dynamically generate features on important interaction points on mesh faces. Experimental results demonstrate that our approach significantly improves the state-of-the-art, especially in scenarios that require accurate gravity or collision reasoning.

Taming the Tail in Class-Conditional GANs: Knowledge Sharing via Unconditional Training at Lower Resolutions

Despite the extensive research on training generative adversarial networks (GANs) with limited training data, learning to generate images from long-tailed training distributions remains fairly unexplored. In the presence of imbalanced multi-class training data, GANs tend to favor classes with more samples, leading to the generation of low-quality and less diverse samples in tail classes. In this study, we aim to improve the training of class-conditional GANs with long-tailed data. We propose a straightforward yet effective method for knowledge sharing, allowing tail classes to borrow from the rich information from classes with more abundant training data. More concretely, we propose modifications to existing class-conditional GAN architectures to ensure that the lower-resolution layers of the generator are trained entirely unconditionally while reserving class-conditional generation for the higher-resolution layers. Experiments on several long-tail benchmarks and GAN architectures demonstrate a significant improvement over existing methods in both the diversity and fidelity of the generated images. The code is available at https://github.com/khorrams/utlo.

Diverse Shape Completion via Style Modulated Generative Adversarial Networks

Shape completion aims to recover the full 3D geometry of an object from a partial observation. This problem is inherently multi-modal since there can be many ways to plausibly complete the missing regions of a shape. Such diversity would be indicative of the underlying uncertainty of the shape and could be preferable for downstream tasks such as planning. In this paper, we propose a novel conditional generative adversarial network that can produce many diverse plausible completions of a partially observed point cloud. To enable our network to produce multiple completions for the same partial input, we introduce stochasticity into our network via style modulation. By extracting style codes from complete shapes during training, and learning a distribution over them, our style codes can explicitly carry shape category information leading to better completions. We further introduce diversity penalties and discriminators at multiple scales to prevent conditional mode collapse and to train without the need for multiple ground truth completions for each partial input. Evaluations across several synthetic and real datasets demonstrate that our method achieves significant improvements in respecting the partial observations while obtaining greater diversity in completions.

Out of Sight, Still in Mind: Reasoning and Planning about Unobserved Objects with Video Tracking Enabled Memory Models

Robots need to have a memory of previously observed, but currently occluded objects to work reliably in realistic environments. We investigate the problem of encoding object-oriented memory into a multi-object manipulation reasoning and planning framework. We propose DOOM and LOOM, which leverage transformer relational dynamics to encode the history of trajectories given partial-view point clouds and an object discovery and tracking engine. Our approaches can perform multiple challenging tasks including reasoning with occluded objects, novel objects appearance, and object reappearance. Throughout our extensive simulation and real-world experiments, we find that our approaches perform well in terms of different numbers of objects and different numbers of distractor actions. Furthermore, we show our approaches outperform an implicit memory baseline.

AutoFocusFormer: Image Segmentation off the Grid

Real world images often have highly imbalanced content density. Some areas are very uniform, e.g., large patches of blue sky, while other areas are scattered with many small objects. Yet, the commonly used successive grid downsampling strategy in convolutional deep networks treats all areas equally. Hence, small objects are represented in very few spatial locations, leading to worse results in tasks such as segmentation. Intuitively, retaining more pixels representing small objects during downsampling helps to preserve important information. To achieve this, we propose AutoFocusFormer (AFF), a local-attention transformer image recognition backbone, which performs adaptive downsampling by learning to retain the most important pixels for the task. Since adaptive downsampling generates a set of pixels irregularly distributed on the image plane, we abandon the classic grid structure. Instead, we develop a novel point-based local attention block, facilitated by a balanced clustering module and a learnable neighborhood merging module, which yields representations for our point-based versions of state-of-the-art segmentation heads. Experiments show that our AutoFocusFormer (AFF) improves significantly over baseline models of similar sizes.