Learning to Ground VLMs without Forgetting

Spatial awareness is key to enable embodied multimodal AI systems. Yet, without vast amounts of spatial supervision, current Visual Language Models (VLMs) struggle at this task. In this paper, we introduce LynX, a framework that equips pretrained VLMs with visual grounding ability without forgetting their existing image and language understanding skills. To this end, we propose a Dual Mixture of Experts module that modifies only the decoder layer of the language model, using one frozen Mixture of Experts (MoE) pre-trained on image and language understanding and another learnable MoE for new grounding capabilities. This allows the VLM to retain previously learned knowledge and skills, while acquiring what is missing. To train the model effectively, we generate a high-quality synthetic dataset we call SCouT, which mimics human reasoning in visual grounding. This dataset provides rich supervision signals, describing a step-by-step multimodal reasoning process, thereby simplifying the task of visual grounding. We evaluate LynX on several object detection and visual grounding datasets, demonstrating strong performance in object detection, zero-shot localization and grounded reasoning while maintaining its original image and language understanding capabilities on seven standard benchmark datasets.

An Image is Worth More Than 16x16 Patches: Exploring Transformers on Individual Pixels

This work does not introduce a new method. Instead, we present an interesting finding that questions the necessity of the inductive bias -- locality in modern computer vision architectures. Concretely, we find that vanilla Transformers can operate by directly treating each individual pixel as a token and achieve highly performant results. This is substantially different from the popular design in Vision Transformer, which maintains the inductive bias from ConvNets towards local neighborhoods (e.g. by treating each 16x16 patch as a token). We mainly showcase the effectiveness of pixels-as-tokens across three well-studied tasks in computer vision: supervised learning for object classification, self-supervised learning via masked autoencoding, and image generation with diffusion models. Although directly operating on individual pixels is less computationally practical, we believe the community must be aware of this surprising piece of knowledge when devising the next generation of neural architectures for computer vision.

Auto-Vocabulary Segmentation for LiDAR Points

Existing perception methods for autonomous driving fall short of recognizing unknown entities not covered in the training data. Open-vocabulary methods offer promising capabilities in detecting any object but are limited by user-specified queries representing target classes. We propose AutoVoc3D, a framework for automatic object class recognition and open-ended segmentation. Evaluation on nuScenes showcases AutoVoc3D's ability to generate precise semantic classes and accurate point-wise segmentation. Moreover, we introduce Text-Point Semantic Similarity, a new metric to assess the semantic similarity between text and point cloud without eliminating novel classes.

Splat-SLAM: Globally Optimized RGB-only SLAM with 3D Gaussians

3D Gaussian Splatting has emerged as a powerful representation of geometry and appearance for RGB-only dense Simultaneous Localization and Mapping (SLAM), as it provides a compact dense map representation while enabling efficient and high-quality map rendering. However, existing methods show significantly worse reconstruction quality than competing methods using other 3D representations, e.g. neural points clouds, since they either do not employ global map and pose optimization or make use of monocular depth. In response, we propose the first RGB-only SLAM system with a dense 3D Gaussian map representation that utilizes all benefits of globally optimized tracking by adapting dynamically to keyframe pose and depth updates by actively deforming the 3D Gaussian map. Moreover, we find that refining the depth updates in inaccurate areas with a monocular depth estimator further improves the accuracy of the 3D reconstruction. Our experiments on the Replica, TUM-RGBD, and ScanNet datasets indicate the effectiveness of globally optimized 3D Gaussians, as the approach achieves superior or on par performance with existing RGB-only SLAM methods methods in tracking, mapping and rendering accuracy while yielding small map sizes and fast runtimes. The source code is available at https://github.com/eriksandstroem/Splat-SLAM.

GlORIE-SLAM: Globally Optimized RGB-only Implicit Encoding Point Cloud SLAM

Recent advancements in RGB-only dense Simultaneous Localization and Mapping (SLAM) have predominantly utilized grid-based neural implicit encodings and/or struggle to efficiently realize global map and pose consistency. To this end, we propose an efficient RGB-only dense SLAM system using a flexible neural point cloud scene representation that adapts to keyframe poses and depth updates, without needing costly backpropagation. Another critical challenge of RGB-only SLAM is the lack of geometric priors. To alleviate this issue, with the aid of a monocular depth estimator, we introduce a novel DSPO layer for bundle adjustment which optimizes the pose and depth of keyframes along with the scale of the monocular depth. Finally, our system benefits from loop closure and online global bundle adjustment and performs either better or competitive to existing dense neural RGB SLAM methods in tracking, mapping and rendering accuracy on the Replica, TUM-RGBD and ScanNet datasets. The source code will be made available.

How NeRFs and 3D Gaussian Splatting are Reshaping SLAM: a Survey

Over the past two decades, research in the field of Simultaneous Localization and Mapping (SLAM) has undergone a significant evolution, highlighting its critical role in enabling autonomous exploration of unknown environments. This evolution ranges from hand-crafted methods, through the era of deep learning, to more recent developments focused on Neural Radiance Fields (NeRFs) and 3D Gaussian Splatting (3DGS) representations. Recognizing the growing body of research and the absence of a comprehensive survey on the topic, this paper aims to provide the first comprehensive overview of SLAM progress through the lens of the latest advancements in radiance fields. It sheds light on the background, evolutionary path, inherent strengths and limitations, and serves as a fundamental reference to highlight the dynamic progress and specific challenges.

Loopy-SLAM: Dense Neural SLAM with Loop Closures

Neural RGBD SLAM techniques have shown promise in dense Simultaneous Localization And Mapping (SLAM), yet face challenges such as error accumulation during camera tracking resulting in distorted maps. In response, we introduce Loopy-SLAM that globally optimizes poses and the dense 3D model. We use frame-to-model tracking using a data-driven point-based submap generation method and trigger loop closures online by performing global place recognition. Robust pose graph optimization is used to rigidly align the local submaps. As our representation is point based, map corrections can be performed efficiently without the need to store the entire history of input frames used for mapping as typically required by methods employing a grid based mapping structure. Evaluation on the synthetic Replica and real-world TUM-RGBD and ScanNet datasets demonstrate competitive or superior performance in tracking, mapping, and rendering accuracy when compared to existing dense neural RGBD SLAM methods. Project page: notchla.github.io/Loopy-SLAM.

Sat2Scene: 3D Urban Scene Generation from Satellite Images with Diffusion

Directly generating scenes from satellite imagery offers exciting possibilities for integration into applications like games and map services. However, challenges arise from significant view changes and scene scale. Previous efforts mainly focused on image or video generation, lacking exploration into the adaptability of scene generation for arbitrary views. Existing 3D generation works either operate at the object level or are difficult to utilize the geometry obtained from satellite imagery. To overcome these limitations, we propose a novel architecture for direct 3D scene generation by introducing diffusion models into 3D sparse representations and combining them with neural rendering techniques. Specifically, our approach generates texture colors at the point level for a given geometry using a 3D diffusion model first, which is then transformed into a scene representation in a feed-forward manner. The representation can be utilized to render arbitrary views which would excel in both single-frame quality and inter-frame consistency. Experiments in two city-scale datasets show that our model demonstrates proficiency in generating photo-realistic street-view image sequences and cross-view urban scenes from satellite imagery.

NeRFmentation: NeRF-based Augmentation for Monocular Depth Estimation

The capabilities of monocular depth estimation (MDE) models are limited by the availability of sufficient and diverse datasets. In the case of MDE models for autonomous driving, this issue is exacerbated by the linearity of the captured data trajectories. We propose a NeRF-based data augmentation pipeline to introduce synthetic data with more diverse viewing directions into training datasets and demonstrate the benefits of our approach to model performance and robustness. Our data augmentation pipeline, which we call "NeRFmentation", trains NeRFs on each scene in the dataset, filters out subpar NeRFs based on relevant metrics, and uses them to generate synthetic RGB-D images captured from new viewing directions. In this work, we apply our technique in conjunction with three state-of-the-art MDE architectures on the popular autonomous driving dataset KITTI, augmenting its training set of the Eigen split. We evaluate the resulting performance gain on the original test set, a separate popular driving set, and our own synthetic test set.

T-MAE: Temporal Masked Autoencoders for Point Cloud Representation Learning

The scarcity of annotated data in outdoor point cloud segmentation poses a significant obstacle in harnessing the modeling capabilities of advanced networks like transformers. Consequently, scholars have been actively investigating efficacious self-supervised pre-training strategies, e.g. contrasting learning and reconstruction-based pretext tasks. Nevertheless, temporal information, which is inherent in the LiDAR point cloud sequence, is consistently disregarded. To better utilize this property, we propose an effective pre-training strategy, namely Temporal Masked AutoEncoders (T-MAE), which takes as input temporally adjacent frames and learns temporal dependency. A SiamWCA backbone, containing a Siamese encoder and a window-based cross-attention (WCA) module, is established for the two-frame input. Taking into account that the motion of an ego-vehicle alters the illumination angles of the same instance, temporal modeling also serves as a robust and natural data augmentation, enhancing the comprehension of target objects. Moreover, instead of utilizing consecutive frames, it is more cost-effective and powerful by using distant historical frames. SiamWCA is a powerful architecture but heavily relies on annotated data. With our T-MAE pre-training strategy, we achieve the best performance on the Waymo dataset among self-supervised learning methods. Comprehensive experiments are conducted to validate all components of our proposal. Upon acceptance, the source code will be made accessible.