Tao Kong

Xingbin Liu, Jinghao Zhou, Tao Kong et al. · bytedance

Masked autoencoders have become popular training paradigms for self-supervised visual representation learning. These models randomly mask a portion of the input and reconstruct the masked portion according to the target representations. In this paper, we first show that a careful choice of the target representation is unnecessary for learning good representations, since different targets tend to derive similarly behaved models. Driven by this observation, we propose a multi-stage masked distillation pipeline and use a randomly initialized model as the teacher, enabling us to effectively train high-capacity models without any efforts to carefully design target representations. Interestingly, we further explore using teachers of larger capacity, obtaining distilled students with remarkable transferring ability. On different tasks of classification, transfer learning, object detection, and semantic segmentation, the proposed method to perform masked knowledge distillation with bootstrapped teachers (dBOT) outperforms previous self-supervised methods by nontrivial margins. We hope our findings, as well as the proposed method, could motivate people to rethink the roles of target representations in pre-training masked autoencoders.The code and pre-trained models are publicly available at https://github.com/liuxingbin/dbot.

Yan Zeng, Hanbo Zhang, Jiani Zheng et al. · bytedance

Recent advancements in Large Language Models (LLMs) such as GPT4 have displayed exceptional multi-modal capabilities in following open-ended instructions given images. However, the performance of these models heavily relies on design choices such as network structures, training data, and training strategies, and these choices have not been extensively discussed in the literature, making it difficult to quantify progress in this field. To address this issue, this paper presents a systematic and comprehensive study, quantitatively and qualitatively, on training such models. We implement over 20 variants with controlled settings. Concretely, for network structures, we compare different LLM backbones and model designs. For training data, we investigate the impact of data and sampling strategies. For instructions, we explore the influence of diversified prompts on the instruction-following ability of the trained models. For benchmarks, we contribute the first, to our best knowledge, comprehensive evaluation set including both image and video tasks through crowd-sourcing. Based on our findings, we present Lynx, which performs the most accurate multi-modal understanding while keeping the best multi-modal generation ability compared to existing open-sourced GPT4-style models.

Xinghang Li, Minghuan Liu, Hanbo Zhang et al.

Recent progress in vision language foundation models has shown their ability to understand multimodal data and resolve complicated vision language tasks, including robotics manipulation. We seek a straightforward way of making use of existing vision-language models (VLMs) with simple fine-tuning on robotics data. To this end, we derive a simple and novel vision-language manipulation framework, dubbed RoboFlamingo, built upon the open-source VLMs, OpenFlamingo. Unlike prior works, RoboFlamingo utilizes pre-trained VLMs for single-step vision-language comprehension, models sequential history information with an explicit policy head, and is slightly fine-tuned by imitation learning only on language-conditioned manipulation datasets. Such a decomposition provides RoboFlamingo the flexibility for open-loop control and deployment on low-performance platforms. By exceeding the state-of-the-art performance with a large margin on the tested benchmark, we show RoboFlamingo can be an effective and competitive alternative to adapt VLMs to robot control. Our extensive experimental results also reveal several interesting conclusions regarding the behavior of different pre-trained VLMs on manipulation tasks. We believe RoboFlamingo has the potential to be a cost-effective and easy-to-use solution for robotics manipulation, empowering everyone with the ability to fine-tune their own robotics policy.

Yunfei Li, Tao Kong, Lei Li et al. · bytedance, cmu

Can a robot autonomously learn to design and construct a bridge from varying-sized blocks without a blueprint? It is a challenging task with long horizon and sparse reward -- the robot has to figure out physically stable design schemes and feasible actions to manipulate and transport blocks. Due to diverse block sizes, the state space and action trajectories are vast to explore. In this paper, we propose a hierarchical approach for this problem. It consists of a reinforcement-learning designer to propose high-level building instructions and a motion-planning-based action generator to manipulate blocks at the low level. For high-level learning, we develop a novel technique, prioritized memory resetting (PMR) to improve exploration. PMR adaptively resets the state to those most critical configurations from a replay buffer so that the robot can resume training on partial architectures instead of from scratch. Furthermore, we augment PMR with auxiliary training objectives and fine-tune the designer with the locomotion generator. Our experiments in simulation and on a real deployed robotic system demonstrate that it is able to effectively construct bridges with blocks of varying sizes at a high success rate. Demos can be found at https://sites.google.com/view/bridge-pmr.

Guanglin Li, Yifeng Li, Zhichao Ye et al. · bytedance

Empowering autonomous agents with 3D understanding for daily objects is a grand challenge in robotics applications. When exploring in an unknown environment, existing methods for object pose estimation are still not satisfactory due to the diversity of object shapes. In this paper, we propose a novel framework for category-level object shape and pose estimation from a single RGB-D image. To handle the intra-category variation, we adopt a semantic primitive representation that encodes diverse shapes into a unified latent space, which is the key to establish reliable correspondences between observed point clouds and estimated shapes. Then, by using a SIM(3)-invariant shape descriptor, we gracefully decouple the shape and pose of an object, thus supporting latent shape optimization of target objects in arbitrary poses. Extensive experiments show that the proposed method achieves SOTA pose estimation performance and better generalization in the real-world dataset. Code and video are available at https://zju3dv.github.io/gCasp.

Taozheng Yang, Ya Jing, Hongtao Wu et al. · bytedance

In this paper, we present a novel method for mobile manipulators to perform multiple contact-rich manipulation tasks. While learning-based methods have the potential to generate actions in an end-to-end manner, they often suffer from insufficient action accuracy and robustness against noise. On the other hand, classical control-based methods can enhance system robustness, but at the cost of extensive parameter tuning. To address these challenges, we present MOMA-Force, a visual-force imitation method that seamlessly combines representation learning for perception, imitation learning for complex motion generation, and admittance whole-body control for system robustness and controllability. MOMA-Force enables a mobile manipulator to learn multiple complex contact-rich tasks with high success rates and small contact forces. In a real household setting, our method outperforms baseline methods in terms of task success rates. Moreover, our method achieves smaller contact forces and smaller force variances compared to baseline methods without force imitation. Overall, we offer a promising approach for efficient and robust mobile manipulation in the real world. Videos and more details can be found on \url{https://visual-force-imitation.github.io}

Rufeng Zhang, Tao Kong, Weihao Wang et al. · bytedance

It is desirable to enable robots capable of automatic assembly. Structural understanding of object parts plays a crucial role in this task yet remains relatively unexplored. In this paper, we focus on the setting of furniture assembly from a complete set of part geometries, which is essentially a 6-DoF part pose estimation problem. We propose a multi-layer transformer-based framework that involves geometric and relational reasoning between parts to update the part poses iteratively. We carefully design a unique instance encoding to solve the ambiguity between geometrically-similar parts so that all parts can be distinguished. In addition to assembling from scratch, we extend our framework to a new task called in-process part assembly. Analogous to furniture maintenance, it requires robots to continue with unfinished products and assemble the remaining parts into appropriate positions. Our method achieves far more than 10% improvements over the current state-of-the-art in multiple metrics on the public PartNet dataset. Extensive experiments and quantitative comparisons demonstrate the effectiveness of the proposed framework.

Ya Jing, Xuelin Zhu, Xingbin Liu et al. · bytedance

Visual pre-training with large-scale real-world data has made great progress in recent years, showing great potential in robot learning with pixel observations. However, the recipes of visual pre-training for robot manipulation tasks are yet to be built. In this paper, we thoroughly investigate the effects of visual pre-training strategies on robot manipulation tasks from three fundamental perspectives: pre-training datasets, model architectures and training methods. Several significant experimental findings are provided that are beneficial for robot learning. Further, we propose a visual pre-training scheme for robot manipulation termed Vi-PRoM, which combines self-supervised learning and supervised learning. Concretely, the former employs contrastive learning to acquire underlying patterns from large-scale unlabeled data, while the latter aims learning visual semantics and temporal dynamics. Extensive experiments on robot manipulations in various simulation environments and the real robot demonstrate the superiority of the proposed scheme. Videos and more details can be found on \url{https://explore-pretrain-robot.github.io}.

Ya Jing, Tao Kong · bytedance

We are witnessing significant progress on perception models, specifically those trained on large-scale internet images. However, efficiently generalizing these perception models to unseen embodied tasks is insufficiently studied, which will help various relevant applications (e.g., home robots). Unlike static perception methods trained on pre-collected images, the embodied agent can move around in the environment and obtain images of objects from any viewpoints. Therefore, efficiently learning the exploration policy and collection method to gather informative training samples is the key to this task. To do this, we first build a 3D semantic distribution map to train the exploration policy self-supervised by introducing the semantic distribution disagreement and the semantic distribution uncertainty rewards. Note that the map is generated from multi-view observations and can weaken the impact of misidentification from an unfamiliar viewpoint. Our agent is then encouraged to explore the objects with different semantic distributions across viewpoints, or uncertain semantic distributions. With the explored informative trajectories, we propose to select hard samples on trajectories based on the semantic distribution uncertainty to reduce unnecessary observations that can be correctly identified. Experiments show that the perception model fine-tuned with our method outperforms the baselines trained with other exploration policies. Further, we demonstrate the robustness of our method in real-robot experiments.

Duo Zheng, Tao Kong, Ya Jing et al. · bytedance

Reference Expression Generation (REG) and Comprehension (REC) are two highly correlated tasks. Modeling REG and REC simultaneously for utilizing the relation between them is a promising way to improve both. However, the problem of distinct inputs, as well as building connections between them in a single model, brings challenges to the design and training of the joint model. To address the problems, we propose a unified model for REG and REC, named UniRef. It unifies these two tasks with the carefully-designed Image-Region-Text Fusion layer (IRTF), which fuses the image, region and text via the image cross-attention and region cross-attention. Additionally, IRTF could generate pseudo input regions for the REC task to enable a uniform way for sharing the identical representation space across the REC and REG. We further propose Vision-conditioned Masked Language Modeling (VMLM) and Text-Conditioned Region Prediction (TRP) to pre-train UniRef model on multi-granular corpora. The VMLM and TRP are directly related to REG and REC, respectively, but could help each other. We conduct extensive experiments on three benchmark datasets, RefCOCO, RefCOCO+ and RefCOCOg. Experimental results show that our model outperforms previous state-of-the-art methods on both REG and REC.

Leyao Liu, Tao Kong, Minzhao Zhu et al. · bytedance

3D instance segmentation methods often require fully-annotated dense labels for training, which are costly to obtain. In this paper, we present ClickSeg, a novel click-level weakly supervised 3D instance segmentation method that requires one point per instance annotation merely. Such a problem is very challenging due to the extremely limited labels, which has rarely been solved before. We first develop a baseline weakly-supervised training method, which generates pseudo labels for unlabeled data by the model itself. To utilize the property of click-level annotation setting, we further propose a new training framework. Instead of directly using the model inference way, i.e., mean-shift clustering, to generate the pseudo labels, we propose to use k-means with fixed initial seeds: the annotated points. New similarity metrics are further designed for clustering. Experiments on ScanNetV2 and S3DIS datasets show that the proposed ClickSeg surpasses the previous best weakly supervised instance segmentation result by a large margin (e.g., +9.4% mAP on ScanNetV2). Using 0.02% supervision signals merely, ClickSeg achieves $\sim$90% of the accuracy of the fully-supervised counterpart. Meanwhile, it also achieves state-of-the-art semantic segmentation results among weakly supervised methods that use the same annotation settings.

Hang Lai, Jiahang Cao, Jiafeng Xu et al.

Legged locomotion over various terrains is challenging and requires precise perception of the robot and its surroundings from both proprioception and vision. However, learning directly from high-dimensional visual input is often data-inefficient and intricate. To address this issue, traditional methods attempt to learn a teacher policy with access to privileged information first and then learn a student policy to imitate the teacher's behavior with visual input. Despite some progress, this imitation framework prevents the student policy from achieving optimal performance due to the information gap between inputs. Furthermore, the learning process is unnatural since animals intuitively learn to traverse different terrains based on their understanding of the world without privileged knowledge. Inspired by this natural ability, we propose a simple yet effective method, World Model-based Perception (WMP), which builds a world model of the environment and learns a policy based on the world model. We illustrate that though completely trained in simulation, the world model can make accurate predictions of real-world trajectories, thus providing informative signals for the policy controller. Extensive simulated and real-world experiments demonstrate that WMP outperforms state-of-the-art baselines in traversability and robustness. Videos and Code are available at: https://wmp-loco.github.io/.

Peiyan Li, Hongtao Wu, Yan Huang et al.

The robotics community has consistently aimed to achieve generalizable robot manipulation with flexible natural language instructions. One primary challenge is that obtaining robot trajectories fully annotated with both actions and texts is time-consuming and labor-intensive. However, partially-annotated data, such as human activity videos without action labels and robot trajectories without text labels, are much easier to collect. Can we leverage these data to enhance the generalization capabilities of robots? In this paper, we propose GR-MG, a novel method which supports conditioning on a text instruction and a goal image. During training, GR-MG samples goal images from trajectories and conditions on both the text and the goal image or solely on the image when text is not available. During inference, where only the text is provided, GR-MG generates the goal image via a diffusion-based image-editing model and conditions on both the text and the generated image. This approach enables GR-MG to leverage large amounts of partially-annotated data while still using languages to flexibly specify tasks. To generate accurate goal images, we propose a novel progress-guided goal image generation model which injects task progress information into the generation process. In simulation experiments, GR-MG improves the average number of tasks completed in a row of 5 from 3.35 to 4.04. In real-robot experiments, GR-MG is able to perform 58 different tasks and improves the success rate from 68.7\% to 78.1\% and 44.4\% to 60.6\% in simple and generalization settings, respectively. It also outperforms comparing baseline methods in few-shot learning of novel skills. Video demos, code, and checkpoints are available on the project page: https://gr-mg.github.io/.

Xinghang Li, Peiyan Li, Minghuan Liu et al.

Foundation Vision Language Models (VLMs) exhibit strong capabilities in multi-modal representation learning, comprehension, and reasoning. By injecting action components into the VLMs, Vision-Language-Action Models (VLAs) can be naturally formed and also show promising performance. Existing work has demonstrated the effectiveness and generalization of VLAs in multiple scenarios and tasks. Nevertheless, the transfer from VLMs to VLAs is not trivial since existing VLAs differ in their backbones, action-prediction formulations, data distributions, and training recipes. This leads to a missing piece for a systematic understanding of the design choices of VLAs. In this work, we disclose the key factors that significantly influence the performance of VLA and focus on answering three essential design choices: which backbone to select, how to formulate the VLA architectures, and when to add cross-embodiment data. The obtained results convince us firmly to explain why we need VLA and develop a new family of VLAs, RoboVLMs, which require very few manual designs and achieve a new state-of-the-art performance in three simulation tasks and real-world experiments. Through our extensive experiments, which include over 8 VLM backbones, 4 policy architectures, and over 600 distinct designed experiments, we provide a detailed guidebook for the future design of VLAs. In addition to the study, the highly flexible RoboVLMs framework, which supports easy integrations of new VLMs and free combinations of various design choices, is made public to facilitate future research. We open-source all details, including codes, models, datasets, and toolkits, along with detailed training and evaluation recipes at: robovlms.github.io.

Hanbo Zhang, Jie Xu, Yuchen Mo et al.

Ambiguity is ubiquitous in human communication. Previous approaches in Human-Robot Interaction (HRI) have often relied on predefined interaction templates, leading to reduced performance in realistic and open-ended scenarios. To address these issues, we present a large-scale dataset, \invig, for interactive visual grounding under language ambiguity. Our dataset comprises over 520K images accompanied by open-ended goal-oriented disambiguation dialogues, encompassing millions of object instances and corresponding question-answer pairs. Leveraging the \invig dataset, we conduct extensive studies and propose a set of baseline solutions for end-to-end interactive visual disambiguation and grounding, achieving a 45.6\% success rate during validation. To the best of our knowledge, the \invig dataset is the first large-scale dataset for resolving open-ended interactive visual grounding, presenting a practical yet highly challenging benchmark for ambiguity-aware HRI. Codes and datasets are available at: \href{https://openivg.github.io}{https://openivg.github.io}.

Yunfei Li, Xiao Ma, Jiafeng Xu et al.

We present GR-RL, a robotic learning framework that turns a generalist vision-language-action (VLA) policy into a highly capable specialist for long-horizon dexterous manipulation. Assuming the optimality of human demonstrations is core to existing VLA policies. However, we claim that in highly dexterous and precise manipulation tasks, human demonstrations are noisy and suboptimal. GR-RL proposes a multi-stage training pipeline that filters, augments, and reinforces the demonstrations by reinforcement learning. First, GR-RL learns a vision-language-conditioned task progress, filters the demonstration trajectories, and only keeps the transitions that contribute positively to the progress. Specifically, we show that by directly applying offline RL with sparse reward, the resulting $Q$-values can be treated as a robust progress function. Next, we introduce morphological symmetry augmentation that greatly improves the generalization and performance of GR-RL. Lastly, to better align the VLA policy with its deployment behaviors for high-precision control, we perform online RL by learning a latent space noise predictor. With this pipeline, GR-RL is, to our knowledge, the first learning-based policy that can autonomously lace up a shoe by threading shoelaces through multiple eyelets with an 83.3% success rate, a task requiring long-horizon reasoning, millimeter-level precision, and compliant soft-body interaction. We hope GR-RL provides a step toward enabling generalist robot foundations models to specialize into reliable real-world experts.

Wenke Xia, Yichu Yang, Hongtao Wu et al.

Establishing a reliable and iteratively refined robotic system is essential for deploying real-world applications. While Vision-Language-Action (VLA) models are widely recognized as the foundation model for such robotic deployment, their reliance on offline expert demonstrations critically limits their capacity for post-deployment refinement. To mitigate this limitation, we introduce Action Preference Optimization (APO), a method designed to refine VLA models by human-assisted preference alignment gathered through interaction with environments. This method begins with a human-robot collaboration framework for reliable failure correction and interaction trajectory collection through human intervention. However, directly leveraging these interaction trajectories for preference optimization is non-trivial due to the challenges of irreversible robotic actions and token distribution mismatch. To solve this, APO proposes an adaptive reweighting algorithm with binary desirability signals derived from interaction, empowering VLA models effectively suppress failure-prone actions while enhancing corrective action adaptation. Ultimately, APO equips VLA models with the crucial capability to learn from failure, paving the way for their iterative refinement and reliable deployment in dynamic environments. The experiments conducted in simulation and real-world scenarios prove superior generalization and robustness of our human-assisted framework across a variety of manipulation tasks. We believe this work could bring insights for efficient and stable optimization of VLA models through human-robot collaboration. The code and dataset are released at https://github.com/GeWu-Lab/Action-Preference-Optimization

Hongtao Wu, Ya Jing, Chilam Cheang et al.

Generative pre-trained models have demonstrated remarkable effectiveness in language and vision domains by learning useful representations. In this paper, we extend the scope of this effectiveness by showing that visual robot manipulation can significantly benefit from large-scale video generative pre-training. We introduce GR-1, a straightforward GPT-style model designed for multi-task language-conditioned visual robot manipulation. GR-1 takes as inputs a language instruction, a sequence of observation images, and a sequence of robot states. It predicts robot actions as well as future images in an end-to-end manner. Thanks to a flexible design, GR-1 can be seamlessly finetuned on robot data after pre-trained on a large-scale video dataset. We perform extensive experiments on the challenging CALVIN benchmark and a real robot. On CALVIN benchmark, our method outperforms state-of-the-art baseline methods and improves the success rate from 88.9% to 94.9%. In the setting of zero-shot unseen scene generalization, GR-1 improves the success rate from 53.3% to 85.4%. In real robot experiments, GR-1 also outperforms baseline methods and shows strong potentials in generalization to unseen scenes and objects. We provide inaugural evidence that a unified GPT-style transformer, augmented with large-scale video generative pre-training, exhibits remarkable generalization to multi-task visual robot manipulation. Project page: https://GR1-Manipulation.github.io

Feng Wang, Tao Kong, Rufeng Zhang et al.

We present TWIST, a simple and theoretically explainable self-supervised representation learning method by classifying large-scale unlabeled datasets in an end-to-end way. We employ a siamese network terminated by a softmax operation to produce twin class distributions of two augmented images. Without supervision, we enforce the class distributions of different augmentations to be consistent. However, simply minimizing the divergence between augmentations will cause collapsed solutions, i.e., outputting the same class probability distribution for all images. In this case, no information about the input image is left. To solve this problem, we propose to maximize the mutual information between the input and the class predictions. Specifically, we minimize the entropy of the distribution for each sample to make the class prediction for each sample assertive and maximize the entropy of the mean distribution to make the predictions of different samples diverse. In this way, TWIST can naturally avoid the collapsed solutions without specific designs such as asymmetric network, stop-gradient operation, or momentum encoder. As a result, TWIST outperforms state-of-the-art methods on a wide range of tasks. Especially, TWIST performs surprisingly well on semi-supervised learning, achieving 61.2% top-1 accuracy with 1% ImageNet labels using a ResNet-50 as backbone, surpassing previous best results by an absolute improvement of 6.2%. Codes and pre-trained models are given on: https://github.com/bytedance/TWIST

Xinlong Wang, Rufeng Zhang, Chunhua Shen et al.

Compared to many other dense prediction tasks, e.g., semantic segmentation, it is the arbitrary number of instances that has made instance segmentation much more challenging. In order to predict a mask for each instance, mainstream approaches either follow the 'detect-then-segment' strategy (e.g., Mask R-CNN), or predict embedding vectors first then cluster pixels into individual instances. In this paper, we view the task of instance segmentation from a completely new perspective by introducing the notion of "instance categories", which assigns categories to each pixel within an instance according to the instance's location. With this notion, we propose segmenting objects by locations (SOLO), a simple, direct, and fast framework for instance segmentation with strong performance. We derive a few SOLO variants (e.g., Vanilla SOLO, Decoupled SOLO, Dynamic SOLO) following the basic principle. Our method directly maps a raw input image to the desired object categories and instance masks, eliminating the need for the grouping post-processing or the bounding box detection. Our approach achieves state-of-the-art results for instance segmentation in terms of both speed and accuracy, while being considerably simpler than the existing methods. Besides instance segmentation, our method yields state-of-the-art results in object detection (from our mask byproduct) and panoptic segmentation. We further demonstrate the flexibility and high-quality segmentation of SOLO by extending it to perform one-stage instance-level image matting. Code is available at: https://git.io/AdelaiDet

Yukang Chen, Yanwei Li, Tao Kong et al.

We propose Scale-aware AutoAug to learn data augmentation policies for object detection. We define a new scale-aware search space, where both image- and box-level augmentations are designed for maintaining scale invariance. Upon this search space, we propose a new search metric, termed Pareto Scale Balance, to facilitate search with high efficiency. In experiments, Scale-aware AutoAug yields significant and consistent improvement on various object detectors (e.g., RetinaNet, Faster R-CNN, Mask R-CNN, and FCOS), even compared with strong multi-scale training baselines. Our searched augmentation policies are transferable to other datasets and box-level tasks beyond object detection (e.g., instance segmentation and keypoint estimation) to improve performance. The search cost is much less than previous automated augmentation approaches for object detection. It is notable that our searched policies have meaningful patterns, which intuitively provide valuable insight for human data augmentation design. Code and models will be available at https://github.com/Jia-Research-Lab/SA-AutoAug.

Peize Sun, Rufeng Zhang, Yi Jiang et al.

We present Sparse R-CNN, a purely sparse method for object detection in images. Existing works on object detection heavily rely on dense object candidates, such as $k$ anchor boxes pre-defined on all grids of image feature map of size $H\times W$. In our method, however, a fixed sparse set of learned object proposals, total length of $N$, are provided to object recognition head to perform classification and location. By eliminating $HWk$ (up to hundreds of thousands) hand-designed object candidates to $N$ (e.g. 100) learnable proposals, Sparse R-CNN completely avoids all efforts related to object candidates design and many-to-one label assignment. More importantly, final predictions are directly output without non-maximum suppression post-procedure. Sparse R-CNN demonstrates accuracy, run-time and training convergence performance on par with the well-established detector baselines on the challenging COCO dataset, e.g., achieving 45.0 AP in standard $3\times$ training schedule and running at 22 fps using ResNet-50 FPN model. We hope our work could inspire re-thinking the convention of dense prior in object detectors. The code is available at: https://github.com/PeizeSun/SparseR-CNN.

Xinlong Wang, Rufeng Zhang, Chunhua Shen et al.

To date, most existing self-supervised learning methods are designed and optimized for image classification. These pre-trained models can be sub-optimal for dense prediction tasks due to the discrepancy between image-level prediction and pixel-level prediction. To fill this gap, we aim to design an effective, dense self-supervised learning method that directly works at the level of pixels (or local features) by taking into account the correspondence between local features. We present dense contrastive learning, which implements self-supervised learning by optimizing a pairwise contrastive (dis)similarity loss at the pixel level between two views of input images. Compared to the baseline method MoCo-v2, our method introduces negligible computation overhead (only <1% slower), but demonstrates consistently superior performance when transferring to downstream dense prediction tasks including object detection, semantic segmentation and instance segmentation; and outperforms the state-of-the-art methods by a large margin. Specifically, over the strong MoCo-v2 baseline, our method achieves significant improvements of 2.0% AP on PASCAL VOC object detection, 1.1% AP on COCO object detection, 0.9% AP on COCO instance segmentation, 3.0% mIoU on PASCAL VOC semantic segmentation and 1.8% mIoU on Cityscapes semantic segmentation. Code is available at: https://git.io/AdelaiDet

Xinlong Wang, Rufeng Zhang, Tao Kong et al.

In this work, we aim at building a simple, direct, and fast instance segmentation framework with strong performance. We follow the principle of the SOLO method of Wang et al. "SOLO: segmenting objects by locations". Importantly, we take one step further by dynamically learning the mask head of the object segmenter such that the mask head is conditioned on the location. Specifically, the mask branch is decoupled into a mask kernel branch and mask feature branch, which are responsible for learning the convolution kernel and the convolved features respectively. Moreover, we propose Matrix NMS (non maximum suppression) to significantly reduce the inference time overhead due to NMS of masks. Our Matrix NMS performs NMS with parallel matrix operations in one shot, and yields better results. We demonstrate a simple direct instance segmentation system, outperforming a few state-of-the-art methods in both speed and accuracy. A light-weight version of SOLOv2 executes at 31.3 FPS and yields 37.1% AP. Moreover, our state-of-the-art results in object detection (from our mask byproduct) and panoptic segmentation show the potential to serve as a new strong baseline for many instance-level recognition tasks besides instance segmentation. Code is available at: https://git.io/AdelaiDet

Luxuan Li, Tao Kong, Fuchun Sun et al.

Detecting actions in videos is an important yet challenging task. Previous works usually utilize (a) sliding window paradigms, or (b) per-frame action scoring and grouping to enumerate the possible temporal locations. Their performances are also limited to the designs of sliding windows or grouping strategies. In this paper, we present a simple and effective method for temporal action proposal generation, named Deep Point-wise Prediction (DPP). DPP simultaneously predicts the action existing possibility and the corresponding temporal locations, without the utilization of any handcrafted sliding window or grouping. The whole system is end-to-end trained with joint loss of temporal action proposal classification and location prediction. We conduct extensive experiments to verify its effectiveness, generality and robustness on standard THUMOS14 dataset. DPP runs more than 1000 frames per second, which largely satisfies the real-time requirement. The code is available at https://github.com/liluxuan1997/DPP.

Xinlong Wang, Wei Yin, Tao Kong et al.

Monocular depth estimation enables 3D perception from a single 2D image, thus attracting much research attention for years. Almost all methods treat foreground and background regions ("things and stuff") in an image equally. However, not all pixels are equal. Depth of foreground objects plays a crucial role in 3D object recognition and localization. To date how to boost the depth prediction accuracy of foreground objects is rarely discussed. In this paper, we first analyse the data distributions and interaction of foreground and background, then propose the foreground-background separated monocular depth estimation (ForeSeE) method, to estimate the foreground depth and background depth using separate optimization objectives and depth decoders. Our method significantly improves the depth estimation performance on foreground objects. Applying ForeSeE to 3D object detection, we achieve 7.5 AP gains and set new state-of-the-art results among other monocular methods. Code will be available at: https://github.com/WXinlong/ForeSeE.

Tao Kong, Fuchun Sun, Huaping Liu et al.

We present FoveaBox, an accurate, flexible, and completely anchor-free framework for object detection. While almost all state-of-the-art object detectors utilize predefined anchors to enumerate possible locations, scales and aspect ratios for the search of the objects, their performance and generalization ability are also limited to the design of anchors. Instead, FoveaBox directly learns the object existing possibility and the bounding box coordinates without anchor reference. This is achieved by: (a) predicting category-sensitive semantic maps for the object existing possibility, and (b) producing category-agnostic bounding box for each position that potentially contains an object. The scales of target boxes are naturally associated with feature pyramid representations. In FoveaBox, an instance is assigned to adjacent feature levels to make the model more accurate.We demonstrate its effectiveness on standard benchmarks and report extensive experimental analysis. Without bells and whistles, FoveaBox achieves state-of-the-art single model performance on the standard COCO and Pascal VOC object detection benchmark. More importantly, FoveaBox avoids all computation and hyper-parameters related to anchor boxes, which are often sensitive to the final detection performance. We believe the simple and effective approach will serve as a solid baseline and help ease future research for object detection. The code has been made publicly available at https://github.com/taokong/FoveaBox .

Chilam Cheang, Sijin Chen, Zhongren Cui et al.

We report our recent progress towards building generalist robot policies, the development of GR-3. GR-3 is a large-scale vision-language-action (VLA) model. It showcases exceptional capabilities in generalizing to novel objects, environments, and instructions involving abstract concepts. Furthermore, it can be efficiently fine-tuned with minimal human trajectory data, enabling rapid and cost-effective adaptation to new settings. GR-3 also excels in handling long-horizon and dexterous tasks, including those requiring bi-manual manipulation and mobile movement, showcasing robust and reliable performance. These capabilities are achieved through a multi-faceted training recipe that includes co-training with web-scale vision-language data, efficient fine-tuning from human trajectory data collected via VR devices, and effective imitation learning with robot trajectory data. In addition, we introduce ByteMini, a versatile bi-manual mobile robot designed with exceptional flexibility and reliability, capable of accomplishing a wide range of tasks when integrated with GR-3. Through extensive real-world experiments, we show GR-3 surpasses the state-of-the-art baseline method, $π_0$, on a wide variety of challenging tasks. We hope GR-3 can serve as a step towards building generalist robots capable of assisting humans in daily life.

Peiyan Li, Yixiang Chen, Hongtao Wu et al.

Recently, leveraging pre-trained vision-language models (VLMs) for building vision-language-action (VLA) models has emerged as a promising approach to effective robot manipulation learning. However, only few methods incorporate 3D signals into VLMs for action prediction, and they do not fully leverage the spatial structure inherent in 3D data, leading to low sample efficiency. In this paper, we introduce BridgeVLA, a novel 3D VLA model that (1) projects 3D inputs to multiple 2D images, ensuring input alignment with the VLM backbone, and (2) utilizes 2D heatmaps for action prediction, unifying the input and output spaces within a consistent 2D image space. In addition, we propose a scalable pre-training method that equips the VLM backbone with the capability to predict 2D heatmaps before downstream policy learning. Extensive experiments show the proposed method is able to learn 3D manipulation efficiently and effectively. BridgeVLA outperforms state-of-the-art baseline methods across three simulation benchmarks. In RLBench, it improves the average success rate from 81.4% to 88.2%. In COLOSSEUM, it demonstrates significantly better performance in challenging generalization settings, boosting the average success rate from 56.7% to 64.0%. In GemBench, it surpasses all the comparing baseline methods in terms of average success rate. In real-robot experiments, BridgeVLA outperforms a state-of-the-art baseline method by 32% on average. It generalizes robustly in multiple out-of-distribution settings, including visual disturbances and unseen instructions. Remarkably, it is able to achieve a success rate of 96.8% on 10+ tasks with only 3 trajectories per task, highlighting its extraordinary sample efficiency. Project Website:https://bridgevla.github.io/

Peiyan Li, Yixiang Chen, Yuan Xu et al.

Robotic manipulation requires understanding both the 3D spatial structure of the environment and its temporal evolution, yet most existing policies overlook one or both. They typically rely on 2D visual observations and backbones pretrained on static image--text pairs, resulting in high data requirements and limited understanding of environment dynamics. To address this, we introduce MV-VDP, a multi-view video diffusion policy that jointly models the 3D spatio-temporal state of the environment. The core idea is to simultaneously predict multi-view heatmap videos and RGB videos, which 1) align the representation format of video pretraining with action finetuning, and 2) specify not only what actions the robot should take, but also how the environment is expected to evolve in response to those actions. Extensive experiments show that MV-VDP enables data-efficient, robust, generalizable, and interpretable manipulation. With only ten demonstration trajectories and without additional pretraining, MV-VDP successfully performs complex real-world tasks, demonstrates strong robustness across a range of model hyperparameters, generalizes to out-of-distribution settings, and predicts realistic future videos. Experiments on Meta-World and real-world robotic platforms demonstrate that MV-VDP consistently outperforms video-prediction--based, 3D-based, and vision--language--action models, establishing a new state of the art in data-efficient multi-task manipulation.

Lei Lv, Yunfei Li, Yu Luo et al.

We present \textbf{FlowRL}, a novel framework for online reinforcement learning that integrates flow-based policy representation with Wasserstein-2-regularized optimization. We argue that in addition to training signals, enhancing the expressiveness of the policy class is crucial for the performance gains in RL. Flow-based generative models offer such potential, excelling at capturing complex, multimodal action distributions. However, their direct application in online RL is challenging due to a fundamental objective mismatch: standard flow training optimizes for static data imitation, while RL requires value-based policy optimization through a dynamic buffer, leading to difficult optimization landscapes. FlowRL first models policies via a state-dependent velocity field, generating actions through deterministic ODE integration from noise. We derive a constrained policy search objective that jointly maximizes Q through the flow policy while bounding the Wasserstein-2 distance to a behavior-optimal policy implicitly derived from the replay buffer. This formulation effectively aligns the flow optimization with the RL objective, enabling efficient and value-aware policy learning despite the complexity of the policy class. Empirical evaluations on DMControl and Humanoidbench demonstrate that FlowRL achieves competitive performance in online reinforcement learning benchmarks.

Minghuan Liu, Zhengbang Zhu, Xiaoshen Han et al.

Modern robotic manipulation primarily relies on visual observations in a 2D color space for skill learning but suffers from poor generalization. In contrast, humans, living in a 3D world, depend more on physical properties-such as distance, size, and shape-than on texture when interacting with objects. Since such 3D geometric information can be acquired from widely available depth cameras, it appears feasible to endow robots with similar perceptual capabilities. Our pilot study found that using depth cameras for manipulation is challenging, primarily due to their limited accuracy and susceptibility to various types of noise. In this work, we propose Camera Depth Models (CDMs) as a simple plugin on daily-use depth cameras, which take RGB images and raw depth signals as input and output denoised, accurate metric depth. To achieve this, we develop a neural data engine that generates high-quality paired data from simulation by modeling a depth camera's noise pattern. Our results show that CDMs achieve nearly simulation-level accuracy in depth prediction, effectively bridging the sim-to-real gap for manipulation tasks. Notably, our experiments demonstrate, for the first time, that a policy trained on raw simulated depth, without the need for adding noise or real-world fine-tuning, generalizes seamlessly to real-world robots on two challenging long-horizon tasks involving articulated, reflective, and slender objects, with little to no performance degradation. We hope our findings will inspire future research in utilizing simulation data and 3D information in general robot policies.

Wenbo Zhang, Tianrun Hu, Yanyuan Qiao et al.

We present Chain-of-Action (CoA), a novel visuo-motor policy paradigm built upon Trajectory Autoregressive Modeling. Unlike conventional approaches that predict next step action(s) forward, CoA generates an entire trajectory by explicit backward reasoning with task-specific goals through an action-level Chain-of-Thought (CoT) process. This process is unified within a single autoregressive structure: (1) the first token corresponds to a stable keyframe action that encodes the task-specific goals; and (2) subsequent action tokens are generated autoregressively, conditioned on the initial keyframe and previously predicted actions. This backward action reasoning enforces a global-to-local structure, allowing each local action to be tightly constrained by the final goal. To further realize the action reasoning structure, CoA incorporates four complementary designs: continuous action token representation; dynamic stopping for variable-length trajectory generation; reverse temporal ensemble; and multi-token prediction to balance action chunk modeling with global structure. As a result, CoA gives strong spatial generalization capabilities while preserving the flexibility and simplicity of a visuo-motor policy. Empirically, we observe CoA achieves the state-of-the-art performance across 60 RLBench tasks and 8 real-world manipulation tasks.

Fangqi Zhu, Hongtao Wu, Song Guo et al.

World models allow autonomous agents to plan and explore by predicting the visual outcomes of different actions. However, for robot manipulation, it is challenging to accurately model the fine-grained robot-object interaction within the visual space using existing methods which overlooks precise alignment between each action and the corresponding frame. In this paper, we present IRASim, a novel world model capable of generating videos with fine-grained robot-object interaction details, conditioned on historical observations and robot action trajectories. We train a diffusion transformer and introduce a novel frame-level action-conditioning module within each transformer block to explicitly model and strengthen the action-frame alignment. Extensive experiments show that: (1) the quality of the videos generated by our method surpasses all the baseline methods and scales effectively with increased model size and computation; (2) policy evaluations using IRASim exhibit a strong correlation with those using the ground-truth simulator, highlighting its potential to accelerate real-world policy evaluation; (3) testing-time scaling through model-based planning with IRASim significantly enhances policy performance, as evidenced by an improvement in the IoU metric on the Push-T benchmark from 0.637 to 0.961; (4) IRASim provides flexible action controllability, allowing virtual robotic arms in datasets to be controlled via a keyboard or VR controller.

Minzhao Zhu, Binglei Zhao, Tao Kong

Object Goal Navigation (ObjectNav) task is to navigate an agent to an object category in unseen environments without a pre-built map. In this paper, we solve this task by predicting the distance to the target using semantically-related objects as cues. Based on the estimated distance to the target object, our method directly choose optimal mid-term goals that are more likely to have a shorter path to the target. Specifically, based on the learned knowledge, our model takes a bird's-eye view semantic map as input, and estimates the path length from the frontier map cells to the target object. With the estimated distance map, the agent could simultaneously explore the environment and navigate to the target objects based on a simple human-designed strategy. Empirical results in visually realistic simulation environments show that the proposed method outperforms a wide range of baselines on success rate and efficiency. Real-robot experiment also demonstrates that our method generalizes well to the real world. Video at https://www.youtube.com/watch?v=R79pWVGFKS4

Jinghao Zhou, Chen Wei, Huiyu Wang et al.

The success of language Transformers is primarily attributed to the pretext task of masked language modeling (MLM), where texts are first tokenized into semantically meaningful pieces. In this work, we study masked image modeling (MIM) and indicate the advantages and challenges of using a semantically meaningful visual tokenizer. We present a self-supervised framework iBOT that can perform masked prediction with an online tokenizer. Specifically, we perform self-distillation on masked patch tokens and take the teacher network as the online tokenizer, along with self-distillation on the class token to acquire visual semantics. The online tokenizer is jointly learnable with the MIM objective and dispenses with a multi-stage training pipeline where the tokenizer needs to be pre-trained beforehand. We show the prominence of iBOT by achieving an 82.3% linear probing accuracy and an 87.8% fine-tuning accuracy evaluated on ImageNet-1K. Beyond the state-of-the-art image classification results, we underline emerging local semantic patterns, which helps the models to obtain strong robustness against common corruptions and achieve leading results on dense downstream tasks, eg., object detection, instance segmentation, and semantic segmentation.

Ruihang Chu, Yukang Chen, Tao Kong et al.

Separating 3D point clouds into individual instances is an important task for 3D vision. It is challenging due to the unknown and varying number of instances in a scene. Existing deep learning based works focus on a two-step pipeline: first learn a feature embedding and then cluster the points. Such a two-step pipeline leads to disconnected intermediate objectives. In this paper, we propose an integrated reformulation of 3D instance segmentation as a per-point classification problem. We propose ICM-3D, a single-step method to segment 3D instances via instantiated categorization. The augmented category information is automatically constructed from 3D spatial positions. We conduct extensive experiments to verify the effectiveness of ICM-3D and show that it obtains inspiring performance across multiple frameworks, backbones and benchmarks.

Yunfei Li, Tao Kong, Lei Li et al.

Autonomous assembly has been a desired functionality of many intelligent robot systems. We study a new challenging assembly task, designing and constructing a bridge without a blueprint. In this task, the robot needs to first design a feasible bridge architecture for arbitrarily wide cliffs and then manipulate the blocks reliably to construct a stable bridge according to the proposed design. In this paper, we propose a bi-level approach to tackle this task. At the high level, the system learns a bridge blueprint policy in a physical simulator using deep reinforcement learning and curriculum learning. A policy is represented as an attention-based neural network with object-centric input, which enables generalization to different numbers of blocks and cliff widths. For low-level control, we implement a motion-planning-based policy for real-robot motion control, which can be directly combined with a trained blueprint policy for real-world bridge construction without tuning. In our field study, our bi-level robot system demonstrates the capability of manipulating blocks to construct a diverse set of bridges with different architectures.

Yiming Li, Tao Kong, Ruihang Chu et al.

Grasping in cluttered scenes has always been a great challenge for robots, due to the requirement of the ability to well understand the scene and object information. Previous works usually assume that the geometry information of the objects is available, or utilize a step-wise, multi-stage strategy to predict the feasible 6-DoF grasp poses. In this work, we propose to formalize the 6-DoF grasp pose estimation as a simultaneous multi-task learning problem. In a unified framework, we jointly predict the feasible 6-DoF grasp poses, instance semantic segmentation, and collision information. The whole framework is jointly optimized and end-to-end differentiable. Our model is evaluated on large-scale benchmarks as well as the real robot system. On the public dataset, our method outperforms prior state-of-the-art methods by a large margin (+4.08 AP). We also demonstrate the implementation of our model on a real robotic platform and show that the robot can accurately grasp target objects in cluttered scenarios with a high success rate. Project link: https://openbyterobotics.github.io/sscl

Mingxuan Jing, Wenbing Huang, Fuchun Sun et al.

It has been a challenge to learning skills for an agent from long-horizon unannotated demonstrations. Existing approaches like Hierarchical Imitation Learning(HIL) are prone to compounding errors or suboptimal solutions. In this paper, we propose Option-GAIL, a novel method to learn skills at long horizon. The key idea of Option-GAIL is modeling the task hierarchy by options and train the policy via generative adversarial optimization. In particular, we propose an Expectation-Maximization(EM)-style algorithm: an E-step that samples the options of expert conditioned on the current learned policy, and an M-step that updates the low- and high-level policies of agent simultaneously to minimize the newly proposed option-occupancy measurement between the expert and the agent. We theoretically prove the convergence of the proposed algorithm. Experiments show that Option-GAIL outperforms other counterparts consistently across a variety of tasks.

Ya Jing, Tao Kong, Wei Wang et al.

Referring image segmentation aims to segment the objects referred by a natural language expression. Previous methods usually focus on designing an implicit and recurrent feature interaction mechanism to fuse the visual-linguistic features to directly generate the final segmentation mask without explicitly modeling the localization information of the referent instances. To tackle these problems, we view this task from another perspective by decoupling it into a "Locate-Then-Segment" (LTS) scheme. Given a language expression, people generally first perform attention to the corresponding target image regions, then generate a fine segmentation mask about the object based on its context. The LTS first extracts and fuses both visual and textual features to get a cross-modal representation, then applies a cross-model interaction on the visual-textual features to locate the referred object with position prior, and finally generates the segmentation result with a light-weight segmentation network. Our LTS is simple but surprisingly effective. On three popular benchmark datasets, the LTS outperforms all the previous state-of-the-art methods by a large margin (e.g., +3.2% on RefCOCO+ and +3.4% on RefCOCOg). In addition, our model is more interpretable with explicitly locating the object, which is also proved by visualization experiments. We believe this framework is promising to serve as a strong baseline for referring image segmentation.

Xinlong Wang, Tao Kong, Chunhua Shen et al.

We present a new, embarrassingly simple approach to instance segmentation in images. Compared to many other dense prediction tasks, e.g., semantic segmentation, it is the arbitrary number of instances that have made instance segmentation much more challenging. In order to predict a mask for each instance, mainstream approaches either follow the 'detect-thensegment' strategy as used by Mask R-CNN, or predict category masks first then use clustering techniques to group pixels into individual instances. We view the task of instance segmentation from a completely new perspective by introducing the notion of "instance categories", which assigns categories to each pixel within an instance according to the instance's location and size, thus nicely converting instance mask segmentation into a classification-solvable problem. Now instance segmentation is decomposed into two classification tasks. We demonstrate a much simpler and flexible instance segmentation framework with strong performance, achieving on par accuracy with Mask R-CNN and outperforming recent singleshot instance segmenters in accuracy. We hope that this very simple and strong framework can serve as a baseline for many instance-level recognition tasks besides instance segmentation.

Chuanqi Tan, Fuchun Sun, Tao Kong et al.

Different functional areas of the human brain play different roles in brain activity, which has not been paid sufficient research attention in the brain-computer interface (BCI) field. This paper presents a new approach for electroencephalography (EEG) classification that applies attention-based transfer learning. Our approach considers the importance of different brain functional areas to improve the accuracy of EEG classification, and provides an additional way to automatically identify brain functional areas associated with new activities without the involvement of a medical professional. We demonstrate empirically that our approach out-performs state-of-the-art approaches in the task of EEG classification, and the results of visualization indicate that our approach can detect brain functional areas related to a certain task.

Tao Kong, Fuchun Sun, Huaping Liu et al.

We present consistent optimization for single stage object detection. Previous works of single stage object detectors usually rely on the regular, dense sampled anchors to generate hypothesis for the optimization of the model. Through an examination of the behavior of the detector, we observe that the misalignment between the optimization target and inference configurations has hindered the performance improvement. We propose to bride this gap by consistent optimization, which is an extension of the traditional single stage detector's optimization strategy. Consistent optimization focuses on matching the training hypotheses and the inference quality by utilizing of the refined anchors during training. To evaluate its effectiveness, we conduct various design choices based on the state-of-the-art RetinaNet detector. We demonstrate it is the consistent optimization, not the architecture design, that yields the performance boosts. Consistent optimization is nearly cost-free, and achieves stable performance gains independent of the model capacities or input scales. Specifically, utilizing consistent optimization improves RetinaNet from 39.1 AP to 40.1 AP on COCO dataset without any bells or whistles, which surpasses the accuracy of all existing state-of-the-art one-stage detectors when adopting ResNet-101 as backbone. The code will be made available.

Liangzhe Yuan, Yibo Chen, Hantian Liu et al.

We propose a light-weight video frame interpolation algorithm. Our key innovation is an instance-level supervision that allows information to be learned from the high-resolution version of similar objects. Our experiment shows that the proposed method can generate state-of-the-art results across different datasets, with fractional computation resources (time and memory) of competing methods. Given two image frames, a cascade network creates an intermediate frame with 1) a flow-warping module that computes coarse bi-directional optical flow and creates an interpolated image via flow-based warping, followed by 2) an image synthesis module to make fine-scale corrections. In the learning stage, object detection proposals are generated on the interpolated image.Lower resolution objects are zoomed into, and the learning algorithms using an adversarial loss trained on high-resolution objects to guide the system towards the instance-level refinement corrects details of object shape and boundaries.

Tao Kong, Fuchun Sun, Wenbing Huang et al.

State-of-the-art object detectors usually learn multi-scale representations to get better results by employing feature pyramids. However, the current designs for feature pyramids are still inefficient to integrate the semantic information over different scales. In this paper, we begin by investigating current feature pyramids solutions, and then reformulate the feature pyramid construction as the feature reconfiguration process. Finally, we propose a novel reconfiguration architecture to combine low-level representations with high-level semantic features in a highly-nonlinear yet efficient way. In particular, our architecture which consists of global attention and local reconfigurations, is able to gather task-oriented features across different spatial locations and scales, globally and locally. Both the global attention and local reconfiguration are lightweight, in-place, and end-to-end trainable. Using this method in the basic SSD system, our models achieve consistent and significant boosts compared with the original model and its other variations, without losing real-time processing speed.

Chuanqi Tan, Fuchun Sun, Tao Kong et al.

As a new classification platform, deep learning has recently received increasing attention from researchers and has been successfully applied to many domains. In some domains, like bioinformatics and robotics, it is very difficult to construct a large-scale well-annotated dataset due to the expense of data acquisition and costly annotation, which limits its development. Transfer learning relaxes the hypothesis that the training data must be independent and identically distributed (i.i.d.) with the test data, which motivates us to use transfer learning to solve the problem of insufficient training data. This survey focuses on reviewing the current researches of transfer learning by using deep neural network and its applications. We defined deep transfer learning, category and review the recent research works based on the techniques used in deep transfer learning.

Tao Kong, Fuchun Sun, Anbang Yao et al.

We present RON, an efficient and effective framework for generic object detection. Our motivation is to smartly associate the best of the region-based (e.g., Faster R-CNN) and region-free (e.g., SSD) methodologies. Under fully convolutional architecture, RON mainly focuses on two fundamental problems: (a) multi-scale object localization and (b) negative sample mining. To address (a), we design the reverse connection, which enables the network to detect objects on multi-levels of CNNs. To deal with (b), we propose the objectness prior to significantly reduce the searching space of objects. We optimize the reverse connection, objectness prior and object detector jointly by a multi-task loss function, thus RON can directly predict final detection results from all locations of various feature maps. Extensive experiments on the challenging PASCAL VOC 2007, PASCAL VOC 2012 and MS COCO benchmarks demonstrate the competitive performance of RON. Specifically, with VGG-16 and low resolution 384X384 input size, the network gets 81.3% mAP on PASCAL VOC 2007, 80.7% mAP on PASCAL VOC 2012 datasets. Its superiority increases when datasets become larger and more difficult, as demonstrated by the results on the MS COCO dataset. With 1.5G GPU memory at test phase, the speed of the network is 15 FPS, 3X faster than the Faster R-CNN counterpart.

Tao Kong, Anbang Yao, Yurong Chen et al.

Almost all of the current top-performing object detection networks employ region proposals to guide the search for object instances. State-of-the-art region proposal methods usually need several thousand proposals to get high recall, thus hurting the detection efficiency. Although the latest Region Proposal Network method gets promising detection accuracy with several hundred proposals, it still struggles in small-size object detection and precise localization (e.g., large IoU thresholds), mainly due to the coarseness of its feature maps. In this paper, we present a deep hierarchical network, namely HyperNet, for handling region proposal generation and object detection jointly. Our HyperNet is primarily based on an elaborately designed Hyper Feature which aggregates hierarchical feature maps first and then compresses them into a uniform space. The Hyper Features well incorporate deep but highly semantic, intermediate but really complementary, and shallow but naturally high-resolution features of the image, thus enabling us to construct HyperNet by sharing them both in generating proposals and detecting objects via an end-to-end joint training strategy. For the deep VGG16 model, our method achieves completely leading recall and state-of-the-art object detection accuracy on PASCAL VOC 2007 and 2012 using only 100 proposals per image. It runs with a speed of 5 fps (including all steps) on a GPU, thus having the potential for real-time processing.