We present Scale3D, a unified framework for scalable 3D reconstruction and scene understanding from a complex and long image sequences. Existing methods typically emphasize either geometric reconstruction or object-level understanding, but struggle to maintain both global geometric consistency and coherent instance identities over hundreds to thousands of views. Our key insight is to exploit their mutual synergy: geometry provides a robust basis for cross-view object association, while perception regularizes and refines geometry. Scale3D decomposes long video into overlapping clusters, reconstructs cluster-wise geometry and 2D segmentation masks, and introduces a 3D-Aware Alignment module to align local predictions into a global proxy geometry while recovering temporally coherent, globally ID-consistent video object segmentation. We further propose Instance-Aware Bundle Adjustment, leveraging dense instance-consistent correspondences to refine the camera poses and geometry. We evaluate Scale3D on ScanNet200 and ScanNet++v2 across three different benchmarking tasks: 3D reconstruction, class-agnostic 3D instance segmentation, and panoptic lifting for novel-view rendering and it achieves the state-of-the-art results with the improvement of 5% on AUC@30, 11% on AP and 10% on Panoptic Quality. Overall, our results highlight the importance of jointly modeling geometry and perception for scalable scene reconstruction and understanding over long image sequences with hundreds to thousands of views.

Monocular visual odometry for unmanned aerial vehicles is challenged by unrestricted 6-DoF motion and large variations in altitude, velocity, scene scale, and frame rate. Existing methods, developed primarily for ground vehicles, often fail to model how geometry and observation time jointly affect aerial motion. We present AeroVO, a lightweight two-frame framework that combines optical flow and monocular geometry to estimate metric linear and angular velocities. AeroVO introduces a scale- and time-aware 3D flow-velocity representation, a multi-scale geometry encoder, and motion-guided fusion for robust ego-motion estimation across diverse flight conditions. We further introduce Reconstructed Aerial Visual Environments (RAVE), a controlled aerial VO dataset containing 140 flight sequences and 21,000 frames rendered from photogrammetric reconstructions of seven real outdoor environments. Experiments on TartanAir and RAVE show that AeroVO outperforms both the evaluated two-frame baselines and a state-of-the-art multiview geometry estimator by 54.7% and 68.7%, respectively, while operating at 30.2 FPS, supporting real-time aerial ego-motion estimation.

We present PanoSeg3R, a feed-forward framework for 3D panoramic semantic segmentation. Unlike existing methods designed for perspective inputs, PanoSeg3R jointly predicts 3D geometry and multi-view semantic segmentation in one single forward pass. Built upon a pretrained reconstruction backbone that supports panoramic images, our approach extends feed-forward 3D reconstruction with a query-based mask decoder. Furthermore, we introduce an automatic panorama data curation pipeline that leverages the complementary strengths of off-the-shelf foundation models to generate reliable pseudo semantic annotations, substantially expanding the training data and improving zero-shot generalization. PanoSeg3R achieves state-of-the-art performance on panoramic 3D semantic segmentation, improving 3D mIoU by up to 16.02 on ScanNet++, while the curated training data further improves zero-shot performance by up to 4.26 and 43.28 mIoU on Stanford2D3D and ToF-360, respectively.

OpenVO effectively estimates real-world–scale ego-motion from monocular dashcam footage with varying observation rates and uncalibrated cameras, enabling robust trajectory dataset construction from rare driving events recorded in dashcam. Existing VO methods are trained on fixed observation frequency (e.g., 10Hz or 12Hz), completely overlooking temporal dynamics information. Many prior methods also require calibrated cameras with known intrinsic parameters. Consequently, their performance degrades when (1) deployed under unseen observation frequencies or (2) applied to uncalibrated cameras. These significantly limit their generalizability to many downstream tasks, such as extracting trajectories from dashcam footage. To address these challenges, OpenVO (1) explicitly encodes temporal dynamics information within a two-frame pose regression framework and (2) leverages 3D geometric priors derived from foundation models. We validate our method on three major autonomous-driving benchmarks -- KITTI, nuScenes, and Argoverse 2 -- achieving more than 20% performance improvement over state-of-the-art approaches. Under varying observation rate settings, our method is significantly more robust, achieving 46%–92% lower errors across all metrics. These results demonstrate the versatility of OpenVO for real-world 3D reconstruction and diverse downstream applications.

Existing 3D instance segmentation methods frequently encounter issues with over-segmentation, leading to redundant and inaccurate 3D proposals that complicate downstream tasks. This challenge arises from their unsupervised merging approach, where dense 2D instance masks are lifted across frames into point clouds to form 3D candidate proposals without direct supervision. These candidates are then hierarchically merged based on heuristic criteria, often resulting in numerous redundant segments that fail to combine into precise 3D proposals. To overcome these limitations, we propose a 3D-Aware 2D Mask Tracking module that uses robust 3D priors from a 2D mask segmentation and tracking foundation model (SAM-2) to ensure consistent object masks across video frames. Rather than merging all visible superpoints across views to create a 3D mask, our 3D Mask Optimization module leverages a dynamic programming algorithm to select an optimal set of views, refining the superpoints to produce a final 3D proposal for each object. Our approach achieves comprehensive object coverage within the scene while reducing unnecessary proposals, which could otherwise impair downstream applications. Evaluations on ScanNet200 and ScanNet++ confirm the effectiveness of our method, with improvements across Class-Agnostic, Open-Vocabulary, and Open-Ended 3D Instance Segmentation tasks.

Open-vocabulary 3D Instance Segmentation methods (OV-3DIS) have recently demonstrated their generalization ability to unseen objects. However, these methods still depend on predefined class names during inference, restricting agents' autonomy. To mitigate this constraint, we propose a novel problem termed Open-Ended 3D Instance Segmentation (OE-3DIS), which eliminates the necessity for predefined class names during testing. We present a comprehensive set of strong baselines inspired by OV-3DIS methodologies, utilizing 2D Multimodal Large Language Models. In addition, we introduce a novel token aggregation strategy that effectively fuses information from multiview images. To evaluate the performance of our OE-3DIS system, we benchmark both the proposed baselines and our method on two widely used indoor datasets: ScanNet200 and ScanNet++. Our approach achieves substantial performance gains over the baselines on both datasets. Notably, even without access to ground-truth object class names during inference, our method outperforms Open3DIS, the current state-of-the-art in OV-3DIS.

Oriented object detection in aerial images poses a significant challenge due to their varying sizes and orientations. Current state-of-the-art detectors typically rely on either two-stage or one-stage approaches, often employing Anchor-based strategies, which can result in computationally expensive operations due to the redundant number of generated anchors during training. In contrast, Anchor-free mechanisms offer faster processing but suffer from a reduction in the number of training samples, potentially impacting detection accuracy. To address these limitations, we propose the Hybrid-Anchor Rotation Detector (HA-RDet), which combines the advantages of both anchor-based and anchor-free schemes for oriented object detection. By utilizing only one preset anchor for each location on the feature maps and refining these anchors with our Orientation-Aware Convolution technique, HA-RDet achieves competitive accuracies, including 75.41 mAP on DOTA-v1, 65.3 mAP on DIOR-R, and 90.2 mAP on HRSC2016, against current anchor-based state-of-the-art methods, while significantly reducing computational resources.

We introduce Open3DIS a novel solution designed to tackle the problem of Open-Vocabulary Instance Segmentation within 3D scenes. Objects within 3D environments exhibit diverse shapes scales and colors making precise instance-level identification a challenging task. Recent advancements in Open-Vocabulary scene understanding have made significant strides in this area by employing class-agnostic 3D instance proposal networks for object localization and learning queryable features for each 3D mask. While these methods produce high-quality instance proposals they struggle with identifying small-scale and geometrically ambiguous objects. The key idea of our method is a new module that aggregates 2D instance masks across frames and maps them to geometrically coherent point cloud regions as high-quality object proposals addressing the above limitations. These are then combined with 3D class-agnostic instance proposals to include a wide range of objects in the real world. To validate our approach we conducted experiments on three prominent datasets including ScanNet200 S3DIS and Replica demonstrating significant performance gains in segmenting objects with diverse categories over the state-of-the-art approaches.