Hong-Yuan Mark Liao

Chien-Yao Wang, Alexey Bochkovskiy, Hong-Yuan Mark Liao · apple-ml

YOLOv7 surpasses all known object detectors in both speed and accuracy in the range from 5 FPS to 160 FPS and has the highest accuracy 56.8% AP among all known real-time object detectors with 30 FPS or higher on GPU V100. YOLOv7-E6 object detector (56 FPS V100, 55.9% AP) outperforms both transformer-based detector SWIN-L Cascade-Mask R-CNN (9.2 FPS A100, 53.9% AP) by 509% in speed and 2% in accuracy, and convolutional-based detector ConvNeXt-XL Cascade-Mask R-CNN (8.6 FPS A100, 55.2% AP) by 551% in speed and 0.7% AP in accuracy, as well as YOLOv7 outperforms: YOLOR, YOLOX, Scaled-YOLOv4, YOLOv5, DETR, Deformable DETR, DINO-5scale-R50, ViT-Adapter-B and many other object detectors in speed and accuracy. Moreover, we train YOLOv7 only on MS COCO dataset from scratch without using any other datasets or pre-trained weights. Source code is released in https://github.com/WongKinYiu/yolov7.

Yu-Hsi Chen, Chien-Yao Wang, Cheng-Yun Yang et al.

We propose a post-processor, called NeighborTrack, that leverages neighbor information of the tracking target to validate and improve single-object tracking (SOT) results. It requires no additional data or retraining. Instead, it uses the confidence score predicted by the backbone SOT network to automatically derive neighbor information and then uses this information to improve the tracking results. When tracking an occluded target, its appearance features are untrustworthy. However, a general siamese network often cannot tell whether the tracked object is occluded by reading the confidence score alone, because it could be misled by neighbors with high confidence scores. Our proposed NeighborTrack takes advantage of unoccluded neighbors' information to reconfirm the tracking target and reduces false tracking when the target is occluded. It not only reduces the impact caused by occlusion, but also fixes tracking problems caused by object appearance changes. NeighborTrack is agnostic to SOT networks and post-processing methods. For the VOT challenge dataset commonly used in short-term object tracking, we improve three famous SOT networks, Ocean, TransT, and OSTrack, by an average of ${1.92\%}$ EAO and ${2.11\%}$ robustness. For the mid- and long-term tracking experiments based on OSTrack, we achieve state-of-the-art ${72.25\%}$ AUC on LaSOT and ${75.7\%}$ AO on GOT-10K. Code duplication can be found in https://github.com/franktpmvu/NeighborTrack.

Wen-Li Wei, Jen-Chun Lin, Tyng-Luh Liu et al.

Learning to capture human motion is essential to 3D human pose and shape estimation from monocular video. However, the existing methods mainly rely on recurrent or convolutional operation to model such temporal information, which limits the ability to capture non-local context relations of human motion. To address this problem, we propose a motion pose and shape network (MPS-Net) to effectively capture humans in motion to estimate accurate and temporally coherent 3D human pose and shape from a video. Specifically, we first propose a motion continuity attention (MoCA) module that leverages visual cues observed from human motion to adaptively recalibrate the range that needs attention in the sequence to better capture the motion continuity dependencies. Then, we develop a hierarchical attentive feature integration (HAFI) module to effectively combine adjacent past and future feature representations to strengthen temporal correlation and refine the feature representation of the current frame. By coupling the MoCA and HAFI modules, the proposed MPS-Net excels in estimating 3D human pose and shape in the video. Though conceptually simple, our MPS-Net not only outperforms the state-of-the-art methods on the 3DPW, MPI-INF-3DHP, and Human3.6M benchmark datasets, but also uses fewer network parameters. The video demos can be found at https://mps-net.github.io/MPS-Net/.

Chien-Yao Wang, Hong-Yuan Mark Liao, I-Hau Yeh

Designing a high-efficiency and high-quality expressive network architecture has always been the most important research topic in the field of deep learning. Most of today's network design strategies focus on how to integrate features extracted from different layers, and how to design computing units to effectively extract these features, thereby enhancing the expressiveness of the network. This paper proposes a new network design strategy, i.e., to design the network architecture based on gradient path analysis. On the whole, most of today's mainstream network design strategies are based on feed forward path, that is, the network architecture is designed based on the data path. In this paper, we hope to enhance the expressive ability of the trained model by improving the network learning ability. Due to the mechanism driving the network parameter learning is the backward propagation algorithm, we design network design strategies based on back propagation path. We propose the gradient path design strategies for the layer-level, the stage-level, and the network-level, and the design strategies are proved to be superior and feasible from theoretical analysis and experiments.

Chien-Yao Wang, I-Hau Yeh, Hong-Yuan Mark Liao

Today's deep learning methods focus on how to design the most appropriate objective functions so that the prediction results of the model can be closest to the ground truth. Meanwhile, an appropriate architecture that can facilitate acquisition of enough information for prediction has to be designed. Existing methods ignore a fact that when input data undergoes layer-by-layer feature extraction and spatial transformation, large amount of information will be lost. This paper will delve into the important issues of data loss when data is transmitted through deep networks, namely information bottleneck and reversible functions. We proposed the concept of programmable gradient information (PGI) to cope with the various changes required by deep networks to achieve multiple objectives. PGI can provide complete input information for the target task to calculate objective function, so that reliable gradient information can be obtained to update network weights. In addition, a new lightweight network architecture -- Generalized Efficient Layer Aggregation Network (GELAN), based on gradient path planning is designed. GELAN's architecture confirms that PGI has gained superior results on lightweight models. We verified the proposed GELAN and PGI on MS COCO dataset based object detection. The results show that GELAN only uses conventional convolution operators to achieve better parameter utilization than the state-of-the-art methods developed based on depth-wise convolution. PGI can be used for variety of models from lightweight to large. It can be used to obtain complete information, so that train-from-scratch models can achieve better results than state-of-the-art models pre-trained using large datasets, the comparison results are shown in Figure 1. The source codes are at: https://github.com/WongKinYiu/yolov9.

Chien-Yao Wang, Hong-Yuan Mark Liao

This is a comprehensive review of the YOLO series of systems. Different from previous literature surveys, this review article re-examines the characteristics of the YOLO series from the latest technical point of view. At the same time, we also analyzed how the YOLO series continued to influence and promote real-time computer vision-related research and led to the subsequent development of computer vision and language models.We take a closer look at how the methods proposed by the YOLO series in the past ten years have affected the development of subsequent technologies and show the applications of YOLO in various fields. We hope this article can play a good guiding role in subsequent real-time computer vision development.

Hung-Shuo Chang, Yue-Cheng Yang, Yu-Hsi Chen et al.

Analyzing animal and human behavior has long been a challenging task in computer vision. Early approaches from the 1970s to the 1990s relied on hand-crafted edge detection, segmentation, and low-level features such as color, shape, and texture to locate objects and infer their identities-an inherently ill-posed problem. Behavior analysis in this era typically proceeded by tracking identified objects over time and modeling their trajectories using sparse feature points, which further limited robustness and generalization. A major shift occurred with the introduction of ImageNet by Deng and Li in 2010, which enabled large-scale visual recognition through deep neural networks and effectively served as a comprehensive visual dictionary. This development allowed object recognition to move beyond complex low-level processing toward learned high-level representations. In this work, we follow this paradigm to build a large-scale Universal Action Space (UAS) using existing labeled human-action datasets. We then use this UAS as the foundation for analyzing and categorizing mammalian and chimpanzee behavior datasets. The source code is released on GitHub at https://github.com/franktpmvu/Universal-Action-Space.

Hung-Shuo Chang, Chien-Yao Wang, Richard Robert Wang et al.

Multi-task learning (MTL) aims to learn multiple tasks using a single model and jointly improve all of them assuming generalization and shared semantics. Reducing conflicts between tasks during joint learning is difficult and generally requires careful network design and extremely large models. We propose building on You Only Learn One Representation (YOLOR), a network architecture specifically designed for multitasking. YOLOR leverages both explicit and implicit knowledge, from data observations and learned latents, respectively, to improve a shared representation while minimizing the number of training parameters. However, YOLOR and its follow-up, YOLOv7, only trained two tasks at once. In this paper, we jointly train object detection, instance segmentation, semantic segmentation, and image captioning. We analyze tradeoffs and attempt to maximize sharing of semantic information. Through our architecture and training strategies, we find that our method achieves competitive performance on all tasks while maintaining a low parameter count and without any pre-training. We will release code soon.

Hao-Tang Tsui, Chien-Yao Wang, Hong-Yuan Mark Liao

Identifying and localizing objects within images is a fundamental challenge, and numerous efforts have been made to enhance model accuracy by experimenting with diverse architectures and refining training strategies. Nevertheless, a prevalent limitation in existing models is overemphasizing the current input while ignoring the information from the entire dataset. We introduce an innovative Retriever-Dictionary (RD) module to address this issue. This architecture enables YOLO-based models to efficiently retrieve features from a Dictionary that contains the insight of the dataset, which is built by the knowledge from Visual Models (VM), Large Language Models (LLM), or Visual Language Models (VLM). The flexible RD enables the model to incorporate such explicit knowledge that enhances the ability to benefit multiple tasks, specifically, segmentation, detection, and classification, from pixel to image level. The experiments show that using the RD significantly improves model performance, achieving more than a 3\% increase in mean Average Precision for object detection with less than a 1% increase in model parameters. Beyond 1-stage object detection models, the RD module improves the effectiveness of 2-stage models and DETR-based architectures, such as Faster R-CNN and Deformable DETR. Code is released at https://github.com/henrytsui000/YOLO.

Chien-Yao Wang, I-Hau Yeh, Hong-Yuan Mark Liao

People ``understand'' the world via vision, hearing, tactile, and also the past experience. Human experience can be learned through normal learning (we call it explicit knowledge), or subconsciously (we call it implicit knowledge). These experiences learned through normal learning or subconsciously will be encoded and stored in the brain. Using these abundant experience as a huge database, human beings can effectively process data, even they were unseen beforehand. In this paper, we propose a unified network to encode implicit knowledge and explicit knowledge together, just like the human brain can learn knowledge from normal learning as well as subconsciousness learning. The unified network can generate a unified representation to simultaneously serve various tasks. We can perform kernel space alignment, prediction refinement, and multi-task learning in a convolutional neural network. The results demonstrate that when implicit knowledge is introduced into the neural network, it benefits the performance of all tasks. We further analyze the implicit representation learnt from the proposed unified network, and it shows great capability on catching the physical meaning of different tasks. The source code of this work is at : https://github.com/WongKinYiu/yolor.

Alexey Bochkovskiy, Chien-Yao Wang, Hong-Yuan Mark Liao

There are a huge number of features which are said to improve Convolutional Neural Network (CNN) accuracy. Practical testing of combinations of such features on large datasets, and theoretical justification of the result, is required. Some features operate on certain models exclusively and for certain problems exclusively, or only for small-scale datasets; while some features, such as batch-normalization and residual-connections, are applicable to the majority of models, tasks, and datasets. We assume that such universal features include Weighted-Residual-Connections (WRC), Cross-Stage-Partial-connections (CSP), Cross mini-Batch Normalization (CmBN), Self-adversarial-training (SAT) and Mish-activation. We use new features: WRC, CSP, CmBN, SAT, Mish activation, Mosaic data augmentation, CmBN, DropBlock regularization, and CIoU loss, and combine some of them to achieve state-of-the-art results: 43.5% AP (65.7% AP50) for the MS COCO dataset at a realtime speed of ~65 FPS on Tesla V100. Source code is at https://github.com/AlexeyAB/darknet

Chien-Yao Wang, Hong-Yuan Mark Liao, I-Hau Yeh et al.

Neural networks have enabled state-of-the-art approaches to achieve incredible results on computer vision tasks such as object detection. However, such success greatly relies on costly computation resources, which hinders people with cheap devices from appreciating the advanced technology. In this paper, we propose Cross Stage Partial Network (CSPNet) to mitigate the problem that previous works require heavy inference computations from the network architecture perspective. We attribute the problem to the duplicate gradient information within network optimization. The proposed networks respect the variability of the gradients by integrating feature maps from the beginning and the end of a network stage, which, in our experiments, reduces computations by 20% with equivalent or even superior accuracy on the ImageNet dataset, and significantly outperforms state-of-the-art approaches in terms of AP50 on the MS COCO object detection dataset. The CSPNet is easy to implement and general enough to cope with architectures based on ResNet, ResNeXt, and DenseNet. Source code is at https://github.com/WongKinYiu/CrossStagePartialNetworks.

Yu-Hsi Chen, Wei-Hsin Chen, Chien-Yao Wang et al.

Assessing chronic pain behavior in mice is critical for preclinical studies. However, existing methods mostly rely on manual labeling of behavioral features, and humans lack a clear understanding of which behaviors best represent chronic pain. For this reason, existing methods struggle to accurately capture the insidious and persistent behavioral changes in chronic pain. This study proposes a framework to automatically discover features related to chronic pain without relying on human-defined action labels. Our method uses universal action space projector to automatically extract mouse action features, and avoids the potential bias of human labeling by retaining the rich behavioral information in the original video. In this paper, we also collected a mouse pain behavior dataset that captures the disease progression of both neuropathic and inflammatory pain across multiple time points. Our method achieves 48.41\% accuracy in a 15-class pain classification task, significantly outperforming human experts (21.33\%) and the widely used method B-SOiD (30.52\%). Furthermore, when the classification is simplified to only three categories, i.e., neuropathic pain, inflammatory pain, and no pain, then our method achieves an accuracy of 73.1\%, which is notably higher than that of human experts (48\%) and B-SOiD (58.43\%). Finally, our method revealed differences in drug efficacy for different types of pain on zero-shot Gabapentin drug testing, and the results were consistent with past drug efficacy literature. This study demonstrates the potential clinical application of our method, which can provide new insights into pain research and related drug development.

Chien-Yao Wang, Alexey Bochkovskiy, Hong-Yuan Mark Liao

We show that the YOLOv4 object detection neural network based on the CSP approach, scales both up and down and is applicable to small and large networks while maintaining optimal speed and accuracy. We propose a network scaling approach that modifies not only the depth, width, resolution, but also structure of the network. YOLOv4-large model achieves state-of-the-art results: 55.5% AP (73.4% AP50) for the MS COCO dataset at a speed of ~16 FPS on Tesla V100, while with the test time augmentation, YOLOv4-large achieves 56.0% AP (73.3 AP50). To the best of our knowledge, this is currently the highest accuracy on the COCO dataset among any published work. The YOLOv4-tiny model achieves 22.0% AP (42.0% AP50) at a speed of 443 FPS on RTX 2080Ti, while by using TensorRT, batch size = 4 and FP16-precision the YOLOv4-tiny achieves 1774 FPS.

Yueh-Hua Wu, I-Hau Yeh, David Hu et al.

The goal of this work is to provide a viable solution based on reinforcement learning for traffic signal control problems. Although the state-of-the-art reinforcement learning approaches have yielded great success in a variety of domains, directly applying it to alleviate traffic congestion can be challenging, considering the requirement of high sample efficiency and how training data is gathered. In this work, we address several challenges that we encountered when we attempted to mitigate serious traffic congestion occurring in a metropolitan area. Specifically, we are required to provide a solution that is able to (1) handle the traffic signal control when certain surveillance cameras that retrieve information for reinforcement learning are down, (2) learn from batch data without a traffic simulator, and (3) make control decisions without shared information across intersections. We present a two-stage framework to deal with the above-mentioned situations. The framework can be decomposed into an Evolution Strategies approach that gives a fixed-time traffic signal control schedule and a multi-agent off-policy reinforcement learning that is capable of learning from batch data with the aid of three proposed components, bounded action, batch augmentation, and surrogate reward clipping. Our experiments show that the proposed framework reduces traffic congestion by 36% in terms of waiting time compared with the currently used fixed-time traffic signal plan. Furthermore, the framework requires only 600 queries to a simulator to achieve the result.

Ping-Yang Chen, Jun-Wei Hsieh, Chien-Yao Wang et al.

State-of-the-art (SoTA) models have improved the accuracy of object detection with a large margin via a FP (feature pyramid). FP is a top-down aggregation to collect semantically strong features to improve scale invariance in both two-stage and one-stage detectors. However, this top-down pathway cannot preserve accurate object positions due to the shift-effect of pooling. Thus, the advantage of FP to improve detection accuracy will disappear when more layers are used. The original FP lacks a bottom-up pathway to offset the lost information from lower-layer feature maps. It performs well in large-sized object detection but poor in small-sized object detection. A new structure "residual feature pyramid" is proposed in this paper. It is bidirectional to fuse both deep and shallow features towards more effective and robust detection for both small-sized and large-sized objects. Due to the "residual" nature, it can be easily trained and integrated to different backbones (even deeper or lighter) than other bi-directional methods. One important property of this residual FP is: accuracy improvement is still found even if more layers are adopted. Extensive experiments on VOC and MS COCO datasets showed the proposed method achieved the SoTA results for highly-accurate and efficient object detection..

Guanjun Guo, Hanzi Wang, Yan Yan et al.

Object proposal generation methods have been widely applied to many computer vision tasks. However, existing object proposal generation methods often suffer from the problems of motion blur, low contrast, deformation, etc., when they are applied to video related tasks. In this paper, we propose an effective and highly accurate target-specific object proposal generation (TOPG) method, which takes full advantage of the context information of a video to alleviate these problems. Specifically, we propose to generate target-specific object proposals by integrating the information of two important objectness cues: colors and edges, which are complementary to each other for different challenging environments in the process of generating object proposals. As a result, the recall of the proposed TOPG method is significantly increased. Furthermore, we propose an object proposal ranking strategy to increase the rank accuracy of the generated object proposals. The proposed TOPG method has yielded significant recall gain (about 20%-60% higher) compared with several state-of-the-art object proposal methods on several challenging visual tracking datasets. Then, we apply the proposed TOPG method to the task of visual tracking and propose a TOPG-based tracker (called as TOPGT), where TOPG is used as a sample selection strategy to select a small number of high-quality target candidates from the generated object proposals. Since the object proposals generated by the proposed TOPG cover many hard negative samples and positive samples, these object proposals can not only be used for training an effective classifier, but also be used as target candidates for visual tracking. Experimental results show the superior performance of TOPGT for visual tracking compared with several other state-of-the-art visual trackers (about 3%-11% higher than the winner of the VOT2015 challenge in term of distance precision).

Guanjun Guo, Hanzi Wang, Chunhua Shen et al.

Despite recent progress, computational visual aesthetic is still challenging. Image cropping, which refers to the removal of unwanted scene areas, is an important step to improve the aesthetic quality of an image. However, it is challenging to evaluate whether cropping leads to aesthetically pleasing results because the assessment is typically subjective. In this paper, we propose a novel cascaded cropping regression (CCR) method to perform image cropping by learning the knowledge from professional photographers. The proposed CCR method improves the convergence speed of the cascaded method, which directly uses random-ferns regressors. In addition, a two-step learning strategy is proposed and used in the CCR method to address the problem of lacking labelled cropping data. Specifically, a deep convolutional neural network (CNN) classifier is first trained on large-scale visual aesthetic datasets. The deep CNN model is then designed to extract features from several image cropping datasets, upon which the cropping bounding boxes are predicted by the proposed CCR method. Experimental results on public image cropping datasets demonstrate that the proposed method significantly outperforms several state-of-the-art image cropping methods.

Huan-Cheng Hsu, Ching-Hang Chen, Hsiao-Rong Tyan et al.

Person re-identification (person re-ID) aims at matching target person(s) grabbed from different and non-overlapping camera views. It plays an important role for public safety and has application in various tasks such as, human retrieval, human tracking, and activity analysis. In this paper, we propose a new network architecture called Hierarchical Cross Network (HCN) to perform person re-ID. In addition to the backbone model of a conventional CNN, HCN is equipped with two additional maps called hierarchical cross feature maps. The maps of an HCN are formed by merging layers with different resolutions and semantic levels. With the hierarchical cross feature maps, an HCN can effectively uncover additional semantic features which could not be discovered by a conventional CNN. Although the proposed HCN can discover features with higher semantics, its representation power is still limited. To derive more general representations, we augment the data during the training process by combining multiple datasets. Experiment results show that the proposed method outperformed several state-of-the-art methods.