In this report, we present some experienced improvements to YOLO series, forming a new high-performance detector -- YOLOX. We switch the YOLO detector to an anchor-free manner and conduct other advanced detection techniques, i.e., a decoupled head and the leading label assignment strategy SimOTA to achieve state-of-the-art results across a large scale range of models: For YOLO-Nano with only 0.91M parameters and 1.08G FLOPs, we get 25.3% AP on COCO, surpassing NanoDet by 1.8% AP; for YOLOv3, one of the most widely used detectors in industry, we boost it to 47.3% AP on COCO, outperforming the current best practice by 3.0% AP; for YOLOX-L with roughly the same amount of parameters as YOLOv4-CSP, YOLOv5-L, we achieve 50.0% AP on COCO at a speed of 68.9 FPS on Tesla V100, exceeding YOLOv5-L by 1.8% AP. Further, we won the 1st Place on Streaming Perception Challenge (Workshop on Autonomous Driving at CVPR 2021) using a single YOLOX-L model. We hope this report can provide useful experience for developers and researchers in practical scenes, and we also provide deploy versions with ONNX, TensorRT, NCNN, and Openvino supported. Source code is at https://github.com/Megvii-BaseDetection/YOLOX.
We introduce the Unity Perception package which aims to simplify and accelerate the process of generating synthetic datasets for computer vision tasks by offering an easy-to-use and highly customizable toolset. This open-source package extends the Unity Editor and engine components to generate perfectly annotated examples for several common computer vision tasks. Additionally, it offers an extensible Randomization framework that lets the user quickly construct and configure randomized simulation parameters in order to introduce variation into the generated datasets. We provide an overview of the provided tools and how they work, and demonstrate the value of the generated synthetic datasets by training a 2D object detection model. The model trained with mostly synthetic data outperforms the model trained using only real data.
Object detection plays an important role in self-driving cars for security development. However, mobile systems on self-driving cars with limited computation resources lead to difficulties for object detection. To facilitate this, we propose a compiler-aware neural pruning search framework to achieve high-speed inference on autonomous vehicles for 2D and 3D object detection. The framework automatically searches the pruning scheme and rate for each layer to find a best-suited pruning for optimizing detection accuracy and speed performance under compiler optimization. Our experiments demonstrate that for the first time, the proposed method achieves (close-to) real-time, 55ms and 99ms inference times for YOLOv4 based 2D object detection and PointPillars based 3D detection, respectively, on an off-the-shelf mobile phone with minor (or no) accuracy loss.
In an autonomous driving system, it is essential to recognize vehicles, pedestrians and cyclists from images. Besides the high accuracy of the prediction, the requirement of real-time running brings new challenges for convolutional network models. In this report, we introduce a real-time method to detect the 2D objects from images. We aggregate several popular one-stage object detectors and train the models of variety input strategies independently, to yield better performance for accurate multi-scale detection of each category, especially for small objects. For model acceleration, we leverage TensorRT to optimize the inference time of our detection pipeline. As shown in the leaderboard, our proposed detection framework ranks the 2nd place with 75.00% L1 mAP and 69.72% L2 mAP in the real-time 2D detection track of the Waymo Open Dataset Challenges, while our framework achieves the latency of 45.8ms/frame on an Nvidia Tesla V100 GPU.
Data augmentation is a key component of CNN based image recognition tasks like object detection. However, it is relatively less explored for 3D object detection. Many standard 2D object detection data augmentation techniques do not extend to 3D box. Extension of these data augmentations for 3D object detection requires adaptation of the 3D geometry of the input scene and synthesis of new viewpoints. This requires accurate depth information of the scene which may not be always available. In this paper, we evaluate existing 2D data augmentations and propose two novel augmentations for monocular 3D detection without a requirement for novel view synthesis. We evaluate these augmentations on the RTM3D detection model firstly due to the shorter training times . We obtain a consistent improvement by 4% in the 3D AP (@IoU=0.7) for cars, ~1.8% scores 3D AP (@IoU=0.25) for pedestrians & cyclists, over the baseline on KITTI car detection dataset. We also demonstrate a rigorous evaluation of the mAP scores by re-weighting them to take into account the class imbalance in the KITTI validation dataset.
Detecting and localizing objects in the real 3D space, which plays a crucial role in scene understanding, is particularly challenging given only a monocular image due to the geometric information loss during imagery projection. We propose MonoGRNet for the amodal 3D object detection from a monocular image via geometric reasoning in both the observed 2D projection and the unobserved depth dimension. MonoGRNet decomposes the monocular 3D object detection task into four sub-tasks including 2D object detection, instance-level depth estimation, projected 3D center estimation and local corner regression. The task decomposition significantly facilitates the monocular 3D object detection, allowing the target 3D bounding boxes to be efficiently predicted in a single forward pass, without using object proposals, post-processing or the computationally expensive pixel-level depth estimation utilized by previous methods. In addition, MonoGRNet flexibly adapts to both fully and weakly supervised learning, which improves the feasibility of our framework in diverse settings. Experiments are conducted on KITTI, Cityscapes and MS COCO datasets. Results demonstrate the promising performance of our framework in various scenarios.
In digital signal processing time-frequency transforms are used to analyze time-varying signals with respect to their spectral contents over time. Apart from the commonly used short-time Fourier transform, other methods exist in literature, such as the Wavelet, Stockwell or Wigner-Ville transform. Consequently, engineers working on digital signal processing tasks are often faced with the question which transform is appropriate for a specific application. To address this question, this paper first briefly introduces the different transforms. Then it compares them with respect to the achievable resolution in time and frequency and possible artifacts. Finally, the paper contains a gallery of time-frequency representations of numerous signals from different fields of applications to allow for visual comparison.
In advanced driver assistant systems and autonomous driving, it is crucial to estimate distances between an ego vehicle and target vehicles. Existing inter-vehicle distance estimation methods assume that the ego and target vehicles drive on a same ground plane. In practical driving environments, however, they may drive on different ground planes. This paper proposes an inter-vehicle distance estimation framework that can consider slope changes of a road forward, by estimating road gradients of \emph{both} ego vehicle and target vehicles and using a 2D object detection deep net. Numerical experiments demonstrate that the proposed method significantly improves the distance estimation accuracy and time complexity, compared to deep learning-based depth estimation methods.
Modern 3D object detectors have immensely benefited from the end-to-end learning idea. However, most of them use a post-processing algorithm called Non-Maximal Suppression (NMS) only during inference. While there were attempts to include NMS in the training pipeline for tasks such as 2D object detection, they have been less widely adopted due to a non-mathematical expression of the NMS. In this paper, we present and integrate GrooMeD-NMS -- a novel Grouped Mathematically Differentiable NMS for monocular 3D object detection, such that the network is trained end-to-end with a loss on the boxes after NMS. We first formulate NMS as a matrix operation and then group and mask the boxes in an unsupervised manner to obtain a simple closed-form expression of the NMS. GrooMeD-NMS addresses the mismatch between training and inference pipelines and, therefore, forces the network to select the best 3D box in a differentiable manner. As a result, GrooMeD-NMS achieves state-of-the-art monocular 3D object detection results on the KITTI benchmark dataset performing comparably to monocular video-based methods. Code and models at https://github.com/abhi1kumar/groomed_nms
A novel representationalist theory of consciousness is presented that is grounded in neuroscience and provides a path to artificially conscious computing. Central to the theory are representational affordances of the conscious experience based on the generation of qualia, the fundamental unit of the conscious representation. The current approach is focused on understanding the balance of simulation, situatedness, and structural coherence of artificial conscious representations through converging evidence from neuroscientific and modeling experiments. Representations instantiating a suitable balance of situated and structurally coherent simulation-based qualia are hypothesized to afford the agent the flexibilities required to succeed in rapidly changing environments.
Universal lesion detection from computed tomography (CT) slices is important for comprehensive disease screening. Since each lesion can locate in multiple adjacent slices, 3D context modeling is of great significance for developing automated lesion detection algorithms. In this work, we propose a Modified Pseudo-3D Feature Pyramid Network (MP3D FPN) that leverages depthwise separable convolutional filters and a group transform module (GTM) to efficiently extract 3D context enhanced 2D features for universal lesion detection in CT slices. To facilitate faster convergence, a novel 3D network pre-training method is derived using solely large-scale 2D object detection dataset in the natural image domain. We demonstrate that with the novel pre-training method, the proposed MP3D FPN achieves state-of-the-art detection performance on the DeepLesion dataset (3.48% absolute improvement in the sensitivity of [email omitted]), significantly surpassing the baseline method by up to 6.06% (in [email omitted]) which adopts 2D convolution for 3D context modeling. Moreover, the proposed 3D pre-trained weights can potentially be used to boost the performance of other 3D medical image analysis tasks.
Accurate 3D object detection (3DOD) is crucial for safe navigation of complex environments by autonomous robots. Regressing accurate 3D bounding boxes in cluttered environments based on sparse LiDAR data is however a highly challenging problem. We address this task by exploring recent advances in conditional energy-based models (EBMs) for probabilistic regression. While methods employing EBMs for regression have demonstrated impressive performance on 2D object detection in images, these techniques are not directly applicable to 3D bounding boxes. In this work, we therefore design a differentiable pooling operator for 3D bounding boxes, serving as the core module of our EBM network. We further integrate this general approach into the state-of-the-art 3D object detector SA-SSD. On the KITTI dataset, our proposed approach consistently outperforms the SA-SSD baseline across all 3DOD metrics, demonstrating the potential of EBM-based regression for highly accurate 3DOD. Code is available at https://github.com/fregu856/ebms_3dod.
The 2D object detection in clean images has been a well studied topic, but its vulnerability against adversarial attack is still worrying. Existing work has improved robustness of object detectors by adversarial training, at the same time, the average precision (AP) on clean images drops significantly. In this paper, we propose that using feature alignment of intermediate layer can improve clean AP and robustness in object detection. Further, on the basis of adversarial training, we present two feature alignment modules: Knowledge-Distilled Feature Alignment (KDFA) module and Self-Supervised Feature Alignment (SSFA) module, which can guide the network to generate more effective features. We conduct extensive experiments on PASCAL VOC and MS-COCO datasets to verify the effectiveness of our proposed approach. The code of our experiments is available at https://github.com/grispeut/Feature-Alignment.git.
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.
While 2D object detection has improved significantly over the past, real world applications of computer vision often require an understanding of the 3D layout of a scene. Many recent approaches to 3D detection use LiDAR point clouds for prediction. We propose a method that only uses a single RGB image, thus enabling applications in devices or vehicles that do not have LiDAR sensors. By using an RGB image, we can leverage the maturity and success of recent 2D object detectors, by extending a 2D detector with a 3D detection head. In this paper we discuss different approaches and experiments, including both regression and classification methods, for designing this 3D detection head. Furthermore, we evaluate how subproblems and implementation details impact the overall prediction result. We use the KITTI dataset for training, which consists of street traffic scenes with class labels, 2D bounding boxes and 3D annotations with seven degrees of freedom. Our final architecture is based on Faster R-CNN. The outputs of the convolutional backbone are fixed sized feature maps for every region of interest. Fully connected layers within the network head then propose an object class and perform 2D bounding box regression. We extend the network head by a 3D detection head, which predicts every degree of freedom of a 3D bounding box via classification. We achieve a mean average precision of 47.3% for moderately difficult data, measured at a 3D intersection over union threshold of 70%, as required by the official KITTI benchmark; outperforming previous state-of-the-art single RGB only methods by a large margin.
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.
In this paper we introduce EfficientPose, a new approach for 6D object pose estimation. Our method is highly accurate, efficient and scalable over a wide range of computational resources. Moreover, it can detect the 2D bounding box of multiple objects and instances as well as estimate their full 6D poses in a single shot. This eliminates the significant increase in runtime when dealing with multiple objects other approaches suffer from. These approaches aim to first detect 2D targets, e.g. keypoints, and solve a Perspective-n-Point problem for their 6D pose for each object afterwards. We also propose a novel augmentation method for direct 6D pose estimation approaches to improve performance and generalization, called 6D augmentation. Our approach achieves a new state-of-the-art accuracy of 97.35% in terms of the ADD(-S) metric on the widely-used 6D pose estimation benchmark dataset Linemod using RGB input, while still running end-to-end at over 27 FPS. Through the inherent handling of multiple objects and instances and the fused single shot 2D object detection as well as 6D pose estimation, our approach runs even with multiple objects (eight) end-to-end at over 26 FPS, making it highly attractive to many real world scenarios. Code will be made publicly available at https://github.com/ybkscht/EfficientPose.
In this paper we introduce the Perception for Autonomous Systems (PAZ) software library. PAZ is a hierarchical perception library that allow users to manipulate multiple levels of abstraction in accordance to their requirements or skill level. More specifically, PAZ is divided into three hierarchical levels which we refer to as pipelines, processors, and backends. These abstractions allows users to compose functions in a hierarchical modular scheme that can be applied for preprocessing, data-augmentation, prediction and postprocessing of inputs and outputs of machine learning (ML) models. PAZ uses these abstractions to build reusable training and prediction pipelines for multiple robot perception tasks such as: 2D keypoint estimation, 2D object detection, 3D keypoint discovery, 6D pose estimation, emotion classification, face recognition, instance segmentation, and attention mechanisms.
DETR has been recently proposed to eliminate the need for many hand-designed components in object detection while demonstrating good performance. However, it suffers from slow convergence and limited feature spatial resolution, due to the limitation of Transformer attention modules in processing image feature maps. To mitigate these issues, we proposed Deformable DETR, whose attention modules only attend to a small set of key sampling points around a reference. Deformable DETR can achieve better performance than DETR (especially on small objects) with 10 times less training epochs. Extensive experiments on the COCO benchmark demonstrate the effectiveness of our approach. Code is released at https://github.com/fundamentalvision/Deformable-DETR.
The growing urban complexity demands an efficient algorithm to acquire and process various sensor information from autonomous vehicles. In this paper, we introduce an algorithm to utilize object detection results from the image to adaptively sample and acquire radar data using Compressed Sensing (CS). This novel algorithm is motivated by the hypothesis that with a limited sampling budget, allocating more sampling budget to areas with the object as opposed to a uniform sampling ultimately improves relevant object detection performance. We improve detection performance by dynamically allocating a lower sampling rate to objects such as buses than pedestrians leading to better reconstruction than baseline across areas with objects of interest. We automate the sampling rate allocation using linear programming and show significant time savings while reducing the radar block size by a factor of 2. We also analyze a Binary Permuted Diagonal measurement matrix for radar acquisition which is hardware-efficient and show its performance is similar to Gaussian and Binary Permuted Block Diagonal matrix. Our experiments on the Oxford radar dataset show an effective reconstruction of objects of interest with 10% sampling rate. Finally, we develop a transformer-based 2D object detection network using the NuScenes radar and image data.