Developing AI agents to autonomously manipulate graphical user interfaces is a long challenging task. Recent advances in data scaling law inspire us to train computer-use agents with a scaled instruction set, yet using behavior cloning to train agents still requires immense high-quality trajectories. To meet the scalability need, we designed STEVE, a step verification pipeline for computer-use agent training. First, we establish a large instruction set for computer-use agents and collect trajectory data with some suboptimal agents. GPT-4o is used to verify the correctness of each step in the trajectories based on the screens before and after the action execution, assigning each step with a binary label. Last, we adopt the Kahneman and Tversky Optimization to optimize the agent from the binary stepwise labels. Extensive experiments manifest that our agent outperforms supervised finetuning by leveraging both positive and negative actions within a trajectory. Also, STEVE enables us to train a 7B vision-language model as a computer-use agent, achieving leading performance in the challenging live desktop environment WinAgentArena with great efficiency at a reduced cost. Code and data: https://github.com/FanbinLu/STEVE.

The increasing use of vision-language models (VLMs) in healthcare applications presents great challenges related to hallucinations, in which the models may generate seemingly plausible results that are in fact incorrect. Such hallucinations can jeopardize clinical decision making, potentially harming the diagnosis and treatments. In this work, we propose MedHallTune, a large-scale benchmark designed specifically to evaluate and mitigate hallucinations in medical VLMs. Comprising over 100,000 images and 1,000,000 instruction pairs, MedHallTune includes both hallucination and non-hallucination samples, each with ground-truth annotations. We conduct a comprehensive evaluation of current medical and general VLMs using MedHallTune, assessing their performance across key metrics, including clinical accuracy, relevance, detail level, and risk level. The experimental results show that fine-tuning with MedHallTune successfully improves the ability of several existing models to manage hallucinations and boost their zero-shot performance on downstream visual-question-answering (VQA) tasks, making them more reliable for practical medical applications. Our work contributes to the development of more trustworthy VLMs. Codes and dataset will be available at MedHallTune.

Learning generative models of 3D point clouds is one of the fundamental problems in 3D generative learning. One of the key properties of point clouds is their permutation invariance, i.e., changing the order of points in a point cloud does not change the shape they represent. In this paper, we analyze the recently proposed equivariant OT flows that learn permutation invariant generative models for pointbased molecular data and we show that these models scale poorly on large point clouds. Also, we observe learning (equivariant) OT flows is generally challenging since straightening flow trajectories makes the learned flow model complex at the beginning of the trajectory. To remedy these, we propose not-so-optimal transport flow models that obtain an approximate OT by an offline OT precomputation, enabling an efficient construction of OT pairs for training. During training, we can additionally construct a hybrid coupling by combining our approximate OT and independent coupling to make the target flow models easier to learn. In an extensive empirical study, we show that our proposed model outperforms prior diffusion-and flow-based approaches on a wide range of unconditional generation and shape completion on the ShapeNet benchmark. Generating 3D point clouds is one of the fundamental problems in 3D modeling with applications in shape generation, 3D reconstruction, 3D design, and perception for robotics and autonomous systems. Recently, diffusion models (Sohl-Dickstein et al., 2015; Ho et al., 2020) and flow matching (Lipman et al., 2022) have become the de facto frameworks for learning generative models for 3D point clouds. These frameworks often overlook 3D point cloud permutation invariance, implying the rearrangement of points does not change the shape that they represent. In closely related areas, equivariant optimal transport (OT) flows (Klein et al., 2024; Song et al., 2024) have been recently developed for 3D molecules that can be considered as sets of 3D atom coordinates.

Few-shot Semantic Segmentation (FSS) is a challenging task that utilizes limited support images to segment associated unseen objects in query images. However, recent FSS methods are observed to perform worse, when enlarging the number of shots. As the support set enlarges, existing FSS networks struggle to concentrate on the high-contributed supports and could easily be overwhelmed by the low-contributed supports that could severely impair the mask predictions. In this work, we study this challenging issue, called support dilution, our goal is to recognize, select, preserve, and enhance those high-contributed supports in the raw support pool. Technically, our method contains three novel parts. First, we propose a contribution index, to quantitatively estimate if a high-contributed support dilutes. Second, we develop the Symmetric Correlation (SC) module to preserve and enhance the high-contributed support features, minimizing the distraction by the low-contributed features. Third, we design the Support Image Pruning operation, to retrieve a compact and high quality subset by discarding low-contributed supports. We conduct extensive experiments on two FSS benchmarks, COCO-20i and PASCAL-5i, the segmentation results demonstrate the compelling performance of our solution over state-of-the-art FSS approaches. Besides, we apply our solution for online segmentation and real-world segmentation, convincing segmentation results showing the practical ability of our work for real-world demonstrations.

We present GeoManip, a framework to enable generalist robots to leverage essential conditions derived from object and part relationships, as geometric constraints, for robot manipulation. For example, cutting the carrot requires adhering to a geometric constraint: the blade of the knife should be perpendicular to the carrot's direction. By interpreting these constraints through symbolic language representations and translating them into low-level actions, GeoManip bridges the gap between natural language and robotic execution, enabling greater generalizability across diverse even unseen tasks, objects, and scenarios. Unlike vision-language-action models that require extensive training, operates training-free by utilizing large foundational models: a constraint generation module that predicts stage-specific geometric constraints and a geometry parser that identifies object parts involved in these constraints. A solver then optimizes trajectories to satisfy inferred constraints from task descriptions and the scene. Furthermore, GeoManip learns in-context and provides five appealing human-robot interaction features: on-the-fly policy adaptation, learning from human demonstrations, learning from failure cases, long-horizon action planning, and efficient data collection for imitation learning. Extensive evaluations on both simulations and real-world scenarios demonstrate GeoManip's state-of-the-art performance, with superior out-of-distribution generalization while avoiding costly model training.

Controversial contents largely inundate the Internet, infringing various cultural norms and child protection standards. Traditional Image Content Moderation (ICM) models fall short in producing precise moderation decisions for diverse standards, while recent multimodal large language models (MLLMs), when adopted to general rule-based ICM, often produce classification and explanation results that are inconsistent with human moderators. Aiming at flexible, explainable, and accurate ICM, we design a novel rule-based dataset generation pipeline, decomposing concise human-defined rules and leveraging well-designed multi-stage prompts to enrich short explicit image annotations. Our ICM-Instruct dataset includes detailed moderation explanation and moderation Q-A pairs. Built upon it, we create our ICM-Assistant model in the framework of rule-based ICM, making it readily applicable in real practice. Our ICM-Assistant model demonstrates exceptional performance and flexibility. Specifically, it significantly outperforms existing approaches on various sources, improving both the moderation classification (36.8\% on average) and moderation explanation quality (26.6\% on average) consistently over existing MLLMs. Code/Data is available at https://github.com/zhaoyuzhi/ICM-Assistant.

In large-scale storehouses, precise instance masks are crucial for robotic bin picking but are challenging to obtain. Existing instance segmentation methods typically rely on a tedious process of scene collection, mask annotation, and network fine-tuning for every single Stock Keeping Unit (SKU). This paper presents SKU-Patch, a new patch-guided instance segmentation solution, leveraging only a few image patches for each incoming new SKU to predict accurate and robust masks, without tedious manual effort and model re-training. Technical-wise, we design a novel transformer-based network with (i) a patch-image correlation encoder to capture multi-level image features calibrated by patch information and (ii) a patch-aware transformer decoder with parallel task heads to generate instance masks. Extensive experiments on four storehouse benchmarks manifest that SKU-Patch is able to achieve the best performance over the state-of-the-art methods. Also, SKU-Patch yields an average of nearly 100% grasping success rate on more than 50 unseen SKUs in a robot-aided auto-store logistic pipeline, showing its effectiveness and practicality.

Robotic bin packing is very challenging, especially when considering practical needs such as object variety and packing compactness. This paper presents SDF-Pack, a new approach based on signed distance field (SDF) to model the geometric condition of objects in a container and compute the object placement locations and packing orders for achieving a more compact bin packing. Our method adopts a truncated SDF representation to localize the computation, and based on it, we formulate the SDF minimization heuristic to find optimized placements to compactly pack objects with the existing ones. To further improve space utilization, if the packing sequence is controllable, our method can suggest which object to be packed next. Experimental results on a large variety of everyday objects show that our method can consistently achieve higher packing compactness over 1,000 packing cases, enabling us to pack more objects into the container, compared with the existing heuristics under various packing settings.

Salient object detection is a fundamental yet challenging problem in computer vision, aiming to highlight the most visually distinctive objects or regions in an image. Recent works benefit from the development of fully convolutional neural networks (FCNs) and achieve great success by integrating features from multiple layers of FCNs. However, the integrated features tend to include non-salient regions (due to low level features of the FCN) or lost details of salient objects (due to high level features of the FCN) when producing the saliency maps. In this paper, we develop a novel deep saliency network equipped with recurrently aggregated deep features (RADF) to more accurately detect salient objects from an image by fully exploiting the complementary saliency information captured in different layers. The RADF utilizes the multi-level features integrated from different layers of a FCN to recurrently refine the features at each layer, suppressing the non-salient noise at low-level of the FCN and increasing more salient details into features at high layers. We perform experiments to evaluate the effectiveness of the proposed network on 5 famous saliency detection benchmarks and compare it with 15 state-of-the-art methods. Our method ranks first in 4 of the 5 datasets and second in the left dataset.