Ziqi Pang

Ziqi Pang, Jie Li, Pavel Tokmakov et al.

This work proposes an end-to-end multi-camera 3D multi-object tracking (MOT) framework. It emphasizes spatio-temporal continuity and integrates both past and future reasoning for tracked objects. Thus, we name it "Past-and-Future reasoning for Tracking" (PF-Track). Specifically, our method adapts the "tracking by attention" framework and represents tracked instances coherently over time with object queries. To explicitly use historical cues, our "Past Reasoning" module learns to refine the tracks and enhance the object features by cross-attending to queries from previous frames and other objects. The "Future Reasoning" module digests historical information and predicts robust future trajectories. In the case of long-term occlusions, our method maintains the object positions and enables re-association by integrating motion predictions. On the nuScenes dataset, our method improves AMOTA by a large margin and remarkably reduces ID-Switches by 90% compared to prior approaches, which is an order of magnitude less. The code and models are made available at https://github.com/TRI-ML/PF-Track.

Ziqi Pang, Deva Ramanan, Mengtian Li et al.

Trajectory forecasting is a widely-studied problem for autonomous navigation. However, existing benchmarks evaluate forecasting based on independent snapshots of trajectories, which are not representative of real-world applications that operate on a continuous stream of data. To bridge this gap, we introduce a benchmark that continuously queries future trajectories on streaming data and we refer to it as "streaming forecasting." Our benchmark inherently captures the disappearance and re-appearance of agents, presenting the emergent challenge of forecasting for occluded agents, which is a safety-critical problem yet overlooked by snapshot-based benchmarks. Moreover, forecasting in the context of continuous timestamps naturally asks for temporal coherence between predictions from adjacent timestamps. Based on this benchmark, we further provide solutions and analysis for streaming forecasting. We propose a plug-and-play meta-algorithm called "Predictive Streamer" that can adapt any snapshot-based forecaster into a streaming forecaster. Our algorithm estimates the states of occluded agents by propagating their positions with multi-modal trajectories, and leverages differentiable filters to ensure temporal consistency. Both occlusion reasoning and temporal coherence strategies significantly improve forecasting quality, resulting in 25% smaller endpoint errors for occluded agents and 10-20% smaller fluctuations of trajectories. Our work is intended to generate interest within the community by highlighting the importance of addressing motion forecasting in its intrinsic streaming setting. Code is available at https://github.com/ziqipang/StreamingForecasting.

Ziqi Pang, Ziyang Xie, Yunze Man et al.

This paper reveals that large language models (LLMs), despite being trained solely on textual data, are surprisingly strong encoders for purely visual tasks in the absence of language. Even more intriguingly, this can be achieved by a simple yet previously overlooked strategy -- employing a frozen transformer block from pre-trained LLMs as a constituent encoder layer to directly process visual tokens. Our work pushes the boundaries of leveraging LLMs for computer vision tasks, significantly departing from conventional practices that typically necessitate a multi-modal vision-language setup with associated language prompts, inputs, or outputs. We demonstrate that our approach consistently enhances performance across a diverse range of tasks, encompassing pure 2D and 3D visual recognition tasks (e.g., image and point cloud classification), temporal modeling tasks (e.g., action recognition), non-semantic tasks (e.g., motion forecasting), and multi-modal tasks (e.g., 2D/3D visual question answering and image-text retrieval). Such improvements are a general phenomenon, applicable to various types of LLMs (e.g., LLaMA and OPT) and different LLM transformer blocks. We additionally propose the information filtering hypothesis to explain the effectiveness of pre-trained LLMs in visual encoding -- the pre-trained LLM transformer blocks discern informative visual tokens and further amplify their effect. This hypothesis is empirically supported by the observation that the feature activation, after training with LLM transformer blocks, exhibits a stronger focus on relevant regions. We hope that our work inspires new perspectives on utilizing LLMs and deepening our understanding of their underlying mechanisms. Code is available at https://github.com/ziqipang/LM4VisualEncoding.

Zheyu Zhang, Ziqi Pang, Shixing Chen et al.

Long video understanding is inherently challenging for vision-language models (VLMs) because of the extensive number of frames. With each video frame typically expanding into tens or hundreds of tokens, the limited context length of large language models (LLMs) forces the VLMs to perceive the frames sparsely and lose temporal information. To address this, we explore extreme video token compression towards one token per frame at the final LLM layer. Our key insight is that heuristic-based compression, widely adopted by previous methods, is prone to information loss, and this necessitates supervising LLM layers into learnable and progressive modules for token-level compression (LP-Comp). Such compression enables our VLM to digest 2x-4x more frames with improved performance. To further increase the token efficiency, we investigate frame-level compression, which selects the frames most relevant to the queries via the internal attention scores of the LLM layers, named question-conditioned compression (QC-Comp). As a notable distinction from previous studies, we mitigate the position bias of LLM attention in long contexts, i.e., the over-concentration on the beginning and end of a sequence, by splitting long videos into short segments and employing local attention. Collectively, our combined token-level and frame-level leads to an extreme compression model for long video understanding, named XComp, achieving a significantly larger compression ratio and enabling denser frame sampling. Our XComp is finetuned from VideoChat-Flash with a data-efficient supervised compression tuning stage that only requires 2.5% of the supervised fine-tuning data, yet boosts the accuracy from 42.9% to 46.2% on LVBench and enhances multiple other long video benchmarks.

Zengyi Qin, Jinyuan Chen, Yunze Man et al.

Training computer use agents requires full-featured OS sandboxes with GUI environments, which consume substantial hardware resources as the number of sandboxes scales. Stochastic errors arising from diverse software execution within these sandboxes further demand robust infrastructure design and reliable error recovery. We present OSGym, a scalable OS environment infrastructure for computer use agents, built around these key optimization strategies: (1) Decentralized OS state management, which isolates failures to individual replicas and significantly enhances overall system reliability; (2) Hardware-aware OS replica orchestration, which addresses CPU-bounded scaling bottlenecks and substantially reduces compute overhead; (3) KVM virtualization with copy-on-write disk management, which shares a common bootable disk across VM instances and provisions only instance-specific modifications, reducing physical disk consumption by 88% and increasing disk provisioning speed by 37 times; and (4) Robust container pool with multi-layer fault recovery. Together, these optimizations yield strong scalability and resource efficiency: OSGym manages over a thousand OS replicas under constrained resources, supports parallel trajectory generation at 1420 multi-turn trajectories per minute, and reduces per-replica cost to 0.2-0.3 USD per day, a 90% reduction over standard deployment. Our experiments validate OSGym across end-to-end pipelines for data collection and training for computer use agents. We believe OSGym establishes a new foundation for scalable, general-purpose computer use agent research.

Michael Ofengenden, Yunze Man, Ziqi Pang et al.

We introduce PPTArena, a benchmark for PowerPoint editing that measures reliable modifications to real slides under natural-language instructions. In contrast to image-PDF renderings or text-to-slide generation, PPTArena focuses on in-place editing across 100 decks, 2125 slides, and over 800 targeted edits covering text, charts, tables, animations, and master-level styles. Each case includes a ground-truth deck, a fully specified target outcome, and a dual VLM-as-judge pipeline that separately scores instruction following and visual quality using both structural diffs and slide images. Building on this setting, we propose PPTPilot, a structure-aware slide-editing agent that plans semantic edit sequences, routes between high-level programmatic tools and deterministic XML operations for precise control, and verifies outputs through an iterative plan-edit-check loop against task-specific constraints. In our experiments, PPTPilot outperforms strong proprietary agents and frontier VLM systems by over 10 percentage points on compound, layout-sensitive, and cross-slide edits, with particularly large gains in visual fidelity and deck-wide consistency. Despite these improvements, existing agents still underperform on long-horizon, document-scale tasks in PPTArena, highlighting the remaining challenges in reliable PPT editing.

Ziqi Pang, Yu-Xiong Wang

We propose MR. Video, an agentic long video understanding framework that demonstrates the simple yet effective MapReduce principle for processing long videos: (1) Map: independently and densely perceiving short video clips, and (2) Reduce: jointly aggregating information from all clips. Compared with sequence-to-sequence vision-language models (VLMs), MR. Video performs detailed short video perception without being limited by context length. Compared with existing video agents that typically rely on sequential key segment selection, the Map operation enables simpler and more scalable sequence parallel perception of short video segments. Its Reduce step allows for more comprehensive context aggregation and reasoning, surpassing explicit key segment retrieval. This MapReduce principle is applicable to both VLMs and video agents, and we use LLM agents to validate its effectiveness. In practice, MR. Video employs two MapReduce stages: (A) Captioning: generating captions for short video clips (map), then standardizing repeated characters and objects into shared names (reduce); (B) Analysis: for each user question, analyzing relevant information from individual short videos (map), and integrating them into a final answer (reduce). MR. Video achieves over 10% accuracy improvement on the challenging LVBench compared to state-of-the-art VLMs and video agents. Code is available at: https://github.com/ziqipang/MR-Video

Ziqi Pang, Xin Xu, Yu-Xiong Wang

With the success of image generation, generative diffusion models are increasingly adopted for discriminative tasks, as pixel generation provides a unified perception interface. However, directly repurposing the generative denoising process for discriminative objectives reveals critical gaps rarely addressed previously. Generative models tolerate intermediate sampling errors if the final distribution remains plausible, but discriminative tasks require rigorous accuracy throughout, as evidenced in challenging multi-modal tasks like referring image segmentation. Motivated by this gap, we analyze and enhance alignment between generative diffusion processes and perception tasks, focusing on how perception quality evolves during denoising. We find: (1) earlier denoising steps contribute disproportionately to perception quality, prompting us to propose tailored learning objectives reflecting varying timestep contributions; (2) later denoising steps show unexpected perception degradation, highlighting sensitivity to training-denoising distribution shifts, addressed by our diffusion-tailored data augmentation; and (3) generative processes uniquely enable interactivity, serving as controllable user interfaces adaptable to correctional prompts in multi-round interactions. Our insights significantly improve diffusion-based perception models without architectural changes, achieving state-of-the-art performance on depth estimation, referring image segmentation, and generalist perception tasks. Code available at https://github.com/ziqipang/ADDP.

Aruna Gauba, Irene Pi, Yunze Man et al.

We present AgMMU, a challenging real-world benchmark for evaluating and advancing vision-language models (VLMs) in the knowledge-intensive domain of agriculture. Unlike prior datasets that rely on crowdsourced prompts, AgMMU is distilled from 116,231 authentic dialogues between everyday growers and USDA-authorized Cooperative Extension experts. Through a three-stage pipeline: automated knowledge extraction, QA generation, and human verification, we construct (i) AgMMU, an evaluation set of 746 multiple-choice questions (MCQs) and 746 open-ended questions (OEQs), and (ii) AgBase, a development corpus of 57,079 multimodal facts covering five high-stakes agricultural topics: insect identification, species identification, disease categorization, symptom description, and management instruction. Benchmarking 12 leading VLMs reveals pronounced gaps in fine-grained perception and factual grounding. Open-sourced models trail after proprietary ones by a wide margin. Simple fine-tuning on AgBase boosts open-sourced model performance on challenging OEQs for up to 11.6% on average, narrowing this gap and also motivating future research to propose better strategies in knowledge extraction and distillation from AgBase. We hope AgMMU stimulates research on domain-specific knowledge integration and trustworthy decision support in agriculture AI development.

Lue Fan, Ziqi Pang, Tianyuan Zhang et al.

In LiDAR-based 3D object detection for autonomous driving, the ratio of the object size to input scene size is significantly smaller compared to 2D detection cases. Overlooking this difference, many 3D detectors directly follow the common practice of 2D detectors, which downsample the feature maps even after quantizing the point clouds. In this paper, we start by rethinking how such multi-stride stereotype affects the LiDAR-based 3D object detectors. Our experiments point out that the downsampling operations bring few advantages, and lead to inevitable information loss. To remedy this issue, we propose Single-stride Sparse Transformer (SST) to maintain the original resolution from the beginning to the end of the network. Armed with transformers, our method addresses the problem of insufficient receptive field in single-stride architectures. It also cooperates well with the sparsity of point clouds and naturally avoids expensive computation. Eventually, our SST achieves state-of-the-art results on the large scale Waymo Open Dataset. It is worth mentioning that our method can achieve exciting performance (83.8 LEVEL 1 AP on validation split) on small object (pedestrian) detection due to the characteristic of single stride. Codes will be released at https://github.com/TuSimple/SST

Qitai Wang, Yuntao Chen, Ziqi Pang et al.

Previous online 3D Multi-Object Tracking(3DMOT) methods terminate a tracklet when it is not associated with new detections for a few frames. But if an object just goes dark, like being temporarily occluded by other objects or simply getting out of FOV, terminating a tracklet prematurely will result in an identity switch. We reveal that premature tracklet termination is the main cause of identity switches in modern 3DMOT systems. To address this, we propose Immortal Tracker, a simple tracking system that utilizes trajectory prediction to maintain tracklets for objects gone dark. We employ a simple Kalman filter for trajectory prediction and preserve the tracklet by prediction when the target is not visible. With this method, we can avoid 96% vehicle identity switches resulting from premature tracklet termination. Without any learned parameters, our method achieves a mismatch ratio at the 0.0001 level and competitive MOTA for the vehicle class on the Waymo Open Dataset test set. Our mismatch ratio is tens of times lower than any previously published method. Similar results are reported on nuScenes. We believe the proposed Immortal Tracker can offer a simple yet powerful solution for pushing the limit of 3DMOT. Our code is available at https://github.com/ImmortalTracker/ImmortalTracker.

Ziqi Pang, Zhichao Li, Naiyan Wang

3D multi-object tracking (MOT) has witnessed numerous novel benchmarks and approaches in recent years, especially those under the "tracking-by-detection" paradigm. Despite their progress and usefulness, an in-depth analysis of their strengths and weaknesses is not yet available. In this paper, we summarize current 3D MOT methods into a unified framework by decomposing them into four constituent parts: pre-processing of detection, association, motion model, and life cycle management. We then ascribe the failure cases of existing algorithms to each component and investigate them in detail. Based on the analyses, we propose corresponding improvements which lead to a strong yet simple baseline: SimpleTrack. Comprehensive experimental results on Waymo Open Dataset and nuScenes demonstrate that our final method could achieve new state-of-the-art results with minor modifications. Furthermore, we take additional steps and rethink whether current benchmarks authentically reflect the ability of algorithms for real-world challenges. We delve into the details of existing benchmarks and find some intriguing facts. Finally, we analyze the distribution and causes of remaining failures in \name\ and propose future directions for 3D MOT. Our code is available at https://github.com/TuSimple/SimpleTrack.

Ziqi Pang, Zhichao Li, Naiyan Wang

Estimating the states of surrounding traffic participants stays at the core of autonomous driving. In this paper, we study a novel setting of this problem: model-free single-object tracking (SOT), which takes the object state in the first frame as input, and jointly solves state estimation and tracking in subsequent frames. The main purpose for this new setting is to break the strong limitation of the popular "detection and tracking" scheme in multi-object tracking. Moreover, we notice that shape completion by overlaying the point clouds, which is a by-product of our proposed task, not only improves the performance of state estimation but also has numerous applications. As no benchmark for this task is available so far, we construct a new dataset LiDAR-SOT and corresponding evaluation protocols based on the Waymo Open dataset. We then propose an optimization-based algorithm called SOTracker involving point cloud registration, vehicle shapes, correspondence, and motion priors. Our quantitative and qualitative results prove the effectiveness of our SOTracker and reveal the challenging cases for SOT in point clouds, including the sparsity of LiDAR data, abrupt motion variation, etc. Finally, we also explore how the proposed task and algorithm may benefit other autonomous driving applications, including simulating LiDAR scans, generating motion data, and annotating optical flow. The code and protocols for our benchmark and algorithm are available at https://github.com/TuSimple/LiDAR_SOT/. A video demonstration is at https://www.youtube.com/watch?v=BpHixKs91i8.

Ziqi Pang, Tianyuan Zhang, Fujun Luan et al.

We introduce RandAR, a decoder-only visual autoregressive (AR) model capable of generating images in arbitrary token orders. Unlike previous decoder-only AR models that rely on a predefined generation order, RandAR removes this inductive bias, unlocking new capabilities in decoder-only generation. Our essential design enables random order by inserting a "position instruction token" before each image token to be predicted, representing the spatial location of the next image token. Trained on randomly permuted token sequences -- a more challenging task than fixed-order generation, RandAR achieves comparable performance to its conventional raster-order counterpart. More importantly, decoder-only transformers trained from random orders acquire new capabilities. For the efficiency bottleneck of AR models, RandAR adopts parallel decoding with KV-Cache at inference time, enjoying 2.5x acceleration without sacrificing generation quality. Additionally, RandAR supports inpainting, outpainting and resolution extrapolation in a zero-shot manner. We hope RandAR inspires new directions for decoder-only visual generation models and broadens their applications across diverse scenarios. Our project page is at https://rand-ar.github.io/.

Lang Lin, Xueyang Yu, Ziqi Pang et al.

This paper proposes a novel framework utilizing multi-modal large language models (MLLMs) for referring video object segmentation (RefVOS). Previous MLLM-based methods commonly struggle with the dilemma between "Ref" and "VOS": they either specialize in understanding a few key frames (global reasoning) or tracking objects on continuous frames (local reasoning), and rely on external VOS or frame selectors to mitigate the other end of the challenge. However, our framework GLUS shows that global and local consistency can be unified into a single video segmentation MLLM: a set of sparse "context frames" provides global information, while a stream of continuous "query frames" conducts local object tracking. This is further supported by jointly training the MLLM with a pre-trained VOS memory bank to simultaneously digest short-range and long-range temporal information. To improve the information efficiency within the limited context window of MLLMs, we introduce object contrastive learning to distinguish hard false-positive objects and a self-refined framework to identify crucial frames and perform propagation. By collectively integrating these insights, our GLUS delivers a simple yet effective baseline, achieving new state-of-the-art for MLLMs on the MeViS and Ref-Youtube-VOS benchmark. Our project page is at https://glus-video.github.io/.

Junbao Zhou, Ziqi Pang, Yu-Xiong Wang

With recent video object segmentation (VOS) benchmarks evolving to challenging scenarios, we revisit a simple but overlooked strategy: restricting the size of memory banks. This diverges from the prevalent practice of expanding memory banks to accommodate extensive historical information. Our specially designed "memory deciphering" study offers a pivotal insight underpinning such a strategy: expanding memory banks, while seemingly beneficial, actually increases the difficulty for VOS modules to decode relevant features due to the confusion from redundant information. By restricting memory banks to a limited number of essential frames, we achieve a notable improvement in VOS accuracy. This process balances the importance and freshness of frames to maintain an informative memory bank within a bounded capacity. Additionally, restricted memory banks reduce the training-inference discrepancy in memory lengths compared with continuous expansion. This fosters new opportunities in temporal reasoning and enables us to introduce the previously overlooked "temporal positional embedding." Finally, our insights are embodied in "RMem" ("R" for restricted), a simple yet effective VOS modification that excels at challenging VOS scenarios and establishes new state of the art for object state changes (on the VOST dataset) and long videos (on the Long Videos dataset). Our code and demo are available at https://restricted-memory.github.io/.

Ziyang Xie, Ziqi Pang, Yu-Xiong Wang

While bird's-eye-view (BEV) perception models can be useful for building high-definition maps (HD-Maps) with less human labor, their results are often unreliable and demonstrate noticeable inconsistencies in the predicted HD-Maps from different viewpoints. This is because BEV perception is typically set up in an 'onboard' manner, which restricts the computation and consequently prevents algorithms from reasoning multiple views simultaneously. This paper overcomes these limitations and advocates a more practical 'offboard' HD-Map generation setup that removes the computation constraints, based on the fact that HD-Maps are commonly reusable infrastructures built offline in data centers. To this end, we propose a novel offboard pipeline called MV-Map that capitalizes multi-view consistency and can handle an arbitrary number of frames with the key design of a 'region-centric' framework. In MV-Map, the target HD-Maps are created by aggregating all the frames of onboard predictions, weighted by the confidence scores assigned by an 'uncertainty network'. To further enhance multi-view consistency, we augment the uncertainty network with the global 3D structure optimized by a voxelized neural radiance field (Voxel-NeRF). Extensive experiments on nuScenes show that our MV-Map significantly improves the quality of HD-Maps, further highlighting the importance of offboard methods for HD-Map generation.

Ziqi Pang, Zhiyuan Hu, Pavel Tokmakov et al.

Virtually all of deep learning literature relies on the assumption of large amounts of available training data. Indeed, even the majority of few-shot learning methods rely on a large set of "base classes" for pretraining. This assumption, however, does not always hold. For some tasks, annotating a large number of classes can be infeasible, and even collecting the images themselves can be a challenge in some scenarios. In this paper, we study this problem and call it "Small Data" setting, in contrast to "Big Data". To unlock the full potential of small data, we propose to augment the models with annotations for other related tasks, thus increasing their generalization abilities. In particular, we use the richly annotated scene parsing dataset ADE20K to construct our realistic Long-tail Recognition with Diverse Supervision (LRDS) benchmark by splitting the object categories into head and tail based on their distribution. Following the standard few-shot learning protocol, we use the head classes for representation learning and the tail classes for evaluation. Moreover, we further subsample the head categories and images to generate two novel settings which we call "Scarce-Class" and "Scarce-Image", respectively corresponding to the shortage of samples for rare classes and training images. Finally, we analyze the effect of applying various additional supervision sources under the proposed settings. Our experiments demonstrate that densely labeling a small set of images can indeed largely remedy the small data constraints.