Wenming Tan

Qiming Yang, Kai Zhang, Chaoxiang Lan et al.

Solid results from Transformers have made them prevailing architectures in various natural language and vision tasks. As a default component in Transformers, Layer Normalization (LN) normalizes activations within each token to boost the robustness. However, LN requires on-the-fly statistics calculation in inference as well as division and square root operations, leading to inefficiency on hardware. What is more, replacing LN with other hardware-efficient normalization schemes (e.g., Batch Normalization) results in inferior performance, even collapse in training. We find that this dilemma is caused by abnormal behaviors of activation statistics, including large fluctuations over iterations and extreme outliers across layers. To tackle these issues, we propose Unified Normalization (UN), which can speed up the inference by being fused with other linear operations and achieve comparable performance on par with LN. UN strives to boost performance by calibrating the activation and gradient statistics with a tailored fluctuation smoothing strategy. Meanwhile, an adaptive outlier filtration strategy is applied to avoid collapse in training whose effectiveness is theoretically proved and experimentally verified in this paper. We demonstrate that UN can be an efficient drop-in alternative to LN by conducting extensive experiments on language and vision tasks. Besides, we evaluate the efficiency of our method on GPU. Transformers equipped with UN enjoy about 31% inference speedup and nearly 18% memory reduction. Code will be released at https://github.com/hikvision-research/Unified-Normalization.

Mengze Li, Tianbao Wang, Haoyu Zhang et al.

Natural language spatial video grounding aims to detect the relevant objects in video frames with descriptive sentences as the query. In spite of the great advances, most existing methods rely on dense video frame annotations, which require a tremendous amount of human effort. To achieve effective grounding under a limited annotation budget, we investigate one-shot video grounding, and learn to ground natural language in all video frames with solely one frame labeled, in an end-to-end manner. One major challenge of end-to-end one-shot video grounding is the existence of videos frames that are either irrelevant to the language query or the labeled frames. Another challenge relates to the limited supervision, which might result in ineffective representation learning. To address these challenges, we designed an end-to-end model via Information Tree for One-Shot video grounding (IT-OS). Its key module, the information tree, can eliminate the interference of irrelevant frames based on branch search and branch cropping techniques. In addition, several self-supervised tasks are proposed based on the information tree to improve the representation learning under insufficient labeling. Experiments on the benchmark dataset demonstrate the effectiveness of our model.

Zeliang Zong, Kai Zhang, Zheyang Li et al.

Large Language Models (LLMs) have demonstrated remarkable proficiency in language comprehension and generation; however, their widespread adoption is constrained by substantial bandwidth and computational demands. While pruning and low-rank approximation have each demonstrated promising performance individually, their synergy for LLMs remains underexplored. We introduce \underline{S}ynergistic \underline{S}parse and \underline{L}ow-Rank \underline{C}ompression (SSLC) methods for LLMs, which leverages the strengths of both techniques: low-rank approximation compresses the model by retaining its essential structure with minimal information loss, whereas sparse optimization eliminates non-essential weights, preserving those crucial for generalization. Based on theoretical analysis, we first formulate the low-rank approximation and sparse optimization as a unified problem and solve it by iterative optimization algorithm. Experiments on LLaMA and Qwen2.5 models (7B-70B) show that SSLC, without any additional training steps, consistently surpasses standalone methods, achieving state-of-the-arts results. Notably, SSLC compresses Qwen2.5 by 50\% with no performance drop and achieves at least 1.63$\times$ speedup, offering a practical solution for efficient LLM deployment.

Xiaodong Yu, Dahu Shi, Xing Wei et al.

This paper presents an end-to-end instance segmentation framework, termed SOIT, that Segments Objects with Instance-aware Transformers. Inspired by DETR \cite{carion2020end}, our method views instance segmentation as a direct set prediction problem and effectively removes the need for many hand-crafted components like RoI cropping, one-to-many label assignment, and non-maximum suppression (NMS). In SOIT, multiple queries are learned to directly reason a set of object embeddings of semantic category, bounding-box location, and pixel-wise mask in parallel under the global image context. The class and bounding-box can be easily embedded by a fixed-length vector. The pixel-wise mask, especially, is embedded by a group of parameters to construct a lightweight instance-aware transformer. Afterward, a full-resolution mask is produced by the instance-aware transformer without involving any RoI-based operation. Overall, SOIT introduces a simple single-stage instance segmentation framework that is both RoI- and NMS-free. Experimental results on the MS COCO dataset demonstrate that SOIT outperforms state-of-the-art instance segmentation approaches significantly. Moreover, the joint learning of multiple tasks in a unified query embedding can also substantially improve the detection performance. Code is available at \url{https://github.com/yuxiaodongHRI/SOIT}.

Dahu Shi, Xing Wei, Xiaodong Yu et al.

Multi-person pose estimation is an attractive and challenging task. Existing methods are mostly based on two-stage frameworks, which include top-down and bottom-up methods. Two-stage methods either suffer from high computational redundancy for additional person detectors or they need to group keypoints heuristically after predicting all the instance-agnostic keypoints. The single-stage paradigm aims to simplify the multi-person pose estimation pipeline and receives a lot of attention. However, recent single-stage methods have the limitation of low performance due to the difficulty of regressing various full-body poses from a single feature vector. Different from previous solutions that involve complex heuristic designs, we present a simple yet effective solution by employing instance-aware dynamic networks. Specifically, we propose an instance-aware module to adaptively adjust (part of) the network parameters for each instance. Our solution can significantly increase the capacity and adaptive-ability of the network for recognizing various poses, while maintaining a compact end-to-end trainable pipeline. Extensive experiments on the MS-COCO dataset demonstrate that our method achieves significant improvement over existing single-stage methods, and makes a better balance of accuracy and efficiency compared to the state-of-the-art two-stage approaches. The code and models are available at \url{https://github.com/hikvision-research/opera}.

Hui Jiang, Yunlu Xu, Zhanzhan Cheng et al.

Text recognition is a popular topic for its broad applications. In this work, we excavate the implicit task, character counting within the traditional text recognition, without additional labor annotation cost. The implicit task plays as an auxiliary branch for complementing the sequential recognition. We design a two-branch reciprocal feature learning framework in order to adequately utilize the features from both the tasks. Through exploiting the complementary effect between explicit and implicit tasks, the feature is reliably enhanced. Extensive experiments on 7 benchmarks show the advantages of the proposed methods in both text recognition and the new-built character counting tasks. In addition, it is convenient yet effective to equip with variable networks and tasks. We offer abundant ablation studies, generalizing experiments with deeper understanding on the tasks. Code is available.

Shuo Li, Jiajun Sun, Guodong Zheng et al.

Recently, multimodal large language models (MLLMs) have demonstrated remarkable performance in visual-language tasks. However, the authenticity of the responses generated by MLLMs is often compromised by object hallucinations. We identify that a key cause of these hallucinations is the model's over-susceptibility to specific image frequency features in detecting objects. In this paper, we introduce Multi-Frequency Perturbations (MFP), a simple, cost-effective, and pluggable method that leverages both low-frequency and high-frequency features of images to perturb visual feature representations and explicitly suppress redundant frequency-domain features during inference, thereby mitigating hallucinations. Experimental results demonstrate that our method significantly mitigates object hallucinations across various model architectures. Furthermore, as a training-time method, MFP can be combined with inference-time methods to achieve state-of-the-art performance on the CHAIR benchmark.

Chen Lin, Zheyang Li, Bo Peng et al.

This paper introduces a post-training quantization~(PTQ) method achieving highly efficient Convolutional Neural Network~ (CNN) quantization with high performance. Previous PTQ methods usually reduce compression error via performing layer-by-layer parameters calibration. However, with lower representational ability of extremely compressed parameters (e.g., the bit-width goes less than 4), it is hard to eliminate all the layer-wise errors. This work addresses this issue via proposing a unit-wise feature reconstruction algorithm based on an observation of second order Taylor series expansion of the unit-wise error. It indicates that leveraging the interaction between adjacent layers' parameters could compensate layer-wise errors better. In this paper, we define several adjacent layers as a Basic-Unit, and present a unit-wise post-training algorithm which can minimize quantization error. This method achieves near-original accuracy on ImageNet and COCO when quantizing FP32 models to INT4 and INT3.

Xing Wei, Yuanrui Kang, Jihao Yang et al.

In recent years, significant progress has been made on the research of crowd counting. However, as the challenging scale variations and complex scenes existed in crowds, neither traditional convolution networks nor recent Transformer architectures with fixed-size attention could handle the task well. To address this problem, this paper proposes a scene-adaptive attention network, termed SAANet. First of all, we design a deformable attention in-built Transformer backbone, which learns adaptive feature representations with deformable sampling locations and dynamic attention weights. Then we propose the multi-level feature fusion and count-attentive feature enhancement modules further, to strengthen feature representation under the global image context. The learned representations could attend to the foreground and are adaptive to different scales of crowds. We conduct extensive experiments on four challenging crowd counting benchmarks, demonstrating that our method achieves state-of-the-art performance. Especially, our method currently ranks No.1 on the public leaderboard of the NWPU-Crowd benchmark. We hope our method could be a strong baseline to support future research in crowd counting. The source code will be released to the community.

Gui-Song Xia, Jian Ding, Ming Qian et al.

This report summarizes the results of Learning to Understand Aerial Images (LUAI) 2021 challenge held on ICCV 2021, which focuses on object detection and semantic segmentation in aerial images. Using DOTA-v2.0 and GID-15 datasets, this challenge proposes three tasks for oriented object detection, horizontal object detection, and semantic segmentation of common categories in aerial images. This challenge received a total of 146 registrations on the three tasks. Through the challenge, we hope to draw attention from a wide range of communities and call for more efforts on the problems of learning to understand aerial images.

Liang Qiao, Zaisheng Li, Zhanzhan Cheng et al.

Table structure recognition is a challenging task due to the various structures and complicated cell spanning relations. Previous methods handled the problem starting from elements in different granularities (rows/columns, text regions), which somehow fell into the issues like lossy heuristic rules or neglect of empty cell division. Based on table structure characteristics, we find that obtaining the aligned bounding boxes of text region can effectively maintain the entire relevant range of different cells. However, the aligned bounding boxes are hard to be accurately predicted due to the visual ambiguities. In this paper, we aim to obtain more reliable aligned bounding boxes by fully utilizing the visual information from both text regions in proposed local features and cell relations in global features. Specifically, we propose the framework of Local and Global Pyramid Mask Alignment, which adopts the soft pyramid mask learning mechanism in both the local and global feature maps. It allows the predicted boundaries of bounding boxes to break through the limitation of original proposals. A pyramid mask re-scoring module is then integrated to compromise the local and global information and refine the predicted boundaries. Finally, we propose a robust table structure recovery pipeline to obtain the final structure, in which we also effectively solve the problems of empty cells locating and division. Experimental results show that the proposed method achieves competitive and even new state-of-the-art performance on several public benchmarks.

Pengbo Zhao, Zhenshen Qu, Yingjia Bu et al.

Fast and precise object detection for high-resolution aerial images has been a challenging task over the years. Due to the sharp variations on object scale, rotation, and aspect ratio, most existing methods are inefficient and imprecise. In this paper, we represent the oriented objects by polar method in polar coordinate and propose PolarDet, a fast and accurate one-stage object detector based on that representation. Our detector introduces a sub-pixel center semantic structure to further improve classifying veracity. PolarDet achieves nearly all SOTA performance in aerial object detection tasks with faster inference speed. In detail, our approach obtains the SOTA results on DOTA, UCAS-AOD, HRSC with 76.64\% mAP, 97.01\% mAP, and 90.46\% mAP respectively. Most noticeably, our PolarDet gets the best performance and reaches the fastest speed(32fps) at the UCAS-AOD dataset.

Zehan Zhang, Ming Zhang, Zhidong Liang et al.

3D vehicle detection based on multi-modal fusion is an important task of many applications such as autonomous driving. Although significant progress has been made, we still observe two aspects that need to be further improvement: First, the specific gain that camera images can bring to 3D detection is seldom explored by previous works. Second, many fusion algorithms run slowly, which is essential for applications with high real-time requirements(autonomous driving). To this end, we propose an end-to-end trainable single-stage multi-modal feature adaptive network in this paper, which uses image information to effectively reduce false positive of 3D detection and has a fast detection speed. A multi-modal adaptive feature fusion module based on channel attention mechanism is proposed to enable the network to adaptively use the feature of each modal. Based on the above mechanism, two fusion technologies are proposed to adapt to different usage scenarios: PointAttentionFusion is suitable for filtering simple false positive and faster; DenseAttentionFusion is suitable for filtering more difficult false positive and has better overall performance. Experimental results on the KITTI dataset demonstrate significant improvement in filtering false positive over the approach using only point cloud data. Furthermore, the proposed method can provide competitive results and has the fastest speed compared to the published state-of-the-art multi-modal methods in the KITTI benchmark.

Bo Peng, Wenming Tan, Zheyang Li et al.

In this paper we propose a novel decomposition method based on filter group approximation, which can significantly reduce the redundancy of deep convolutional neural networks (CNNs) while maintaining the majority of feature representation. Unlike other low-rank decomposition algorithms which operate on spatial or channel dimension of filters, our proposed method mainly focuses on exploiting the filter group structure for each layer. For several commonly used CNN models, including VGG and ResNet, our method can reduce over 80% floating-point operations (FLOPs) with less accuracy drop than state-of-the-art methods on various image classification datasets. Besides, experiments demonstrate that our method is conducive to alleviating degeneracy of the compressed network, which hurts the convergence and performance of the network.