BAN: Focusing on Boundary Context for Object Detection

An object detector based on convolutional neural network (CNN) has been widely used in the field of computer vision because of its simplicity and efficiency. The average accuracy of CNN model detection results in the object detector is greatly affected by the loss function. The precision of the localization algorithm in the loss function is the main factor affecting the result. Based on the complete intersection over union (CIoU) loss function, an improved penalty function is proposed to improve the localization accuracy. Specifically, the algorithm more comprehensively considers matching bounding boxes between prediction with ground truth, using the proportional relationship of the aspect ratio from both bounding boxes. Under the same aspect ratio of the two bounding boxes, the influence factors of the prediction box on localization accuracy were considered. In this way, the function of the penalty function is strengthened, and localization accuracy of the network model improved. This loss function is called Improved CIoU (ICIoU). Experiments on the Udacity, PASCAL VOC, and MS COCO datasets have demonstrated the effectiveness of ICIoU in improving localization accuracy of network models by using the one-stage object detector YOLOv4. Compared with CIoU, the proposed ICIoU improved average precision (AP) by 0.57% and AP75 by 0.12% on Udacity, AP by 0.26% and AP75 by 1.28% on PASCAL VOC, and AP by 0.06% and AP75 by 0.65% on MS COCO.

The development of object detection networks has reached a high point, and there have been significant improvements in accuracy and detection speed. Object detection is widely used in intelligent robots, self‐driving cars, and other edge‐intelligent terminals. Unfortunately, when a detector is allowed to learn new objects in an unfamiliar environment, it can catastrophically forget the objects it has already learned. In particular, reliable and stable knowledge cannot be extracted from old models. Based on this, a new multinetwork mean distillation loss function for open‐world domain incremental object detection is presented. To better extract reliable and stable knowledge from old models, we enhanced the distillation output of the detector with a ResNet50 backbone and an output RoI head. The distillation output of the intermediate RPN is softened by adaptive distillation. To obtain more stable results, the ResNet50 backbone and RPN on the channel are zero‐averaged. Various incremental steps and stability experiments are performed on two benchmark datasets, PASCAL VOC and MS COCO. The experimental results show the excellent performance of our method in different experimental scenarios, and it is superior to the most advanced methods. For example, in the setting of the batch task, incremental object detection on the PASCAL VOC and MS COCO datasets is improved by 3.4% and 2.1%, respectively.

In this paper, we propose an open-set object detection framework based on a dynamic hierarchical structure with incremental learning capabilities for unseen object classes. We were motivated by the observation that deep features extracted from visual objects show a strong hierarchical clustering property. The hierarchical feature model (HFM) was used to learn a new object class by using collaborative sampling (CS), and open-set-aware active semi-supervised learning (ASSL) algorithms. We divided object proposals into superclasses by using the agglomerative clustering algorithm. Data samples in each superclass node were classified into multiple augmented class nodes instead of directly associating with regular object classes. One or more augmented class nodes are related to a regular object class, and each augmented class has only one superclass. Object proposals from inexperienced data distribution are assigned to an augmented class node. Dynamic HFM nodes in the decision path are assembled to constitute an ensemble prediction, and the new augmented object is associated with a new regular object class. Our experimental results showed that the proposed method uses standard benchmark datasets such as PASCAL VOC, MS COCO, ILSVRC DET, and local datasets to perform better than state-of-the-art techniques.

How to add more context information and bring more accurate detection is an important problem to be considered in object detection. In this paper, we propose a new object detector with enriched global context information by a pyramid feature pool module and several global activation blocks, named EGCI-Net, which is a one-stage object detector from scratch as DSOD.The global activation blocks are added into the backbone sub network of the detector to weaken the local information of the detected object feature maps and increase the global context of them. And the pyramid feature pool module produces multi-scale global context features to supervise the pyramid features by multi-scale global average pooling. Then the features obtained by the main structure are fused with the pyramid pooling features to merge into the final multibox detector. We have evaluated our detector on the Pascal VOC and MS COCO datasets. The experimental results show that our proposed detector achieves better results than DSOD and exceeds most of the existing excellent detectors, especially detects partially occluded objects and small objects well.

We aim to model the top-down attention of a Convolutional Neural Network (CNN) classifier for generating task-specific attention maps. Inspired by a top-down human visual attention model, we propose a new backpropagation scheme, called Excitation Backprop, to pass along top-down signals downwards in the network hierarchy via a probabilistic Winner-Take-All process. Furthermore, we introduce the concept of contrastive attention to make the top-down attention maps more discriminative. In experiments, we demonstrate the accuracy and generalizability of our method in weakly supervised localization tasks on the MS COCO, PASCAL VOC07 and ImageNet datasets. The usefulness of our method is further validated in the text-to-region association task. On the Flickr30k Entities dataset, we achieve promising performance in phrase localization by leveraging the top-down attention of a CNN model that has been trained on weakly labeled web images.

Object detection has received significant interest in the research field of computer vision and is widely used in human-centric applications. The occlusion problem is a frequent obstacle that degrades detection quality. In this paper, we propose a novel object detection framework targeting robust object detection in occlusion. The proposed deep learning-based network consists mainly of two parts: 1) object detection framework, which classifies the object categories and localizes the object location and 2) plug-in bounding-box (BB) estimator, which estimates the object and occlusion region from the feature map of the backbone network and the corresponding critic network for evaluating the predicted BB map. The BB estimator and the critic network are the plug-in modules added to the object detection framework and learned competitively with adversarial manner. As the plug-in BB estimator is learned to estimate the BB map containing the object and occlusion pattern information, the backbone network can embed this information to enable robust detection under occlusion in the test phase. The comprehensive experimental results on the PASCAL VOC, MS COCO, and KITTI dataset showed that the performance is improved with the plug-in BB-Critic network by predicting and criticizing object and occlusion in general generic object detection framework.

Disentangling the intrinsic feature from the related feature in image classification using knowledge distillation and object replacement

To further improve the speed and accuracy of object detection, especially small targets and occluded objects, a novel and efficient detector named YOLO-ACN is presented. The detector model is inspired by the high detection accuracy and speed of YOLOv3, and it is improved by the addition of an attention mechanism, a CIoU (complete intersection over union) loss function, Soft-NMS (non-maximum suppression), and depthwise separable convolution. First, the attention mechanism is introduced in the channel and spatial dimensions in each residual block to focus on small targets. Second, CIoU loss is adopted to achieve accurate bounding box (BBox) regression. Besides, to filter out a more accurate BBox and avoid deleting occluded objects in dense images, the CIoU is applied in the Soft-NMS, and the Gaussian model in the Soft-NMS is employed to suppress the surrounding BBox. Third, to significantly reduce the parameters and improve the detection speed, standard convolution is replaced by depthwise separable convolution, and hard-swish activation function is utilized in deeper layers. On the MS COCO dataset and infrared pedestrian dataset KAIST, the quantitative experimental results show that compared with other state-of-the-art models, the proposed YOLO-ACN has high accuracy and speed in detecting small targets and occluded objects. YOLO-ACN reaches a mAP50 (mean average precision) of 53.8% and an APs (average precision for small objects) of 18.2% at a real-time speed of 22 ms on the MS COCO dataset, and the mAP for a single class on the KAIST dataset even reaches over 80% on an NVIDIA Tesla K40.

In the systems of industrial robotics and autonomous vehicles, instance segmentation is widely employed. However, manually labelling an object outline is time‐consuming. In order to reduce annotation costs, we present a weakly supervised instance segmentation method in this article. A deeply convolutional network is first used to construct multi‐scale feature maps for each object in the input image. After that, the encoder‐decoder framework with dynamic convolution is utilised to enhance model capacity and efficiency, while avoiding the issues of anchor design, proposal selection, and RoIAlign implementation. In particular, Dynamic Heads are used in the encoder to create dynamic convolution kernels, while Instance Heads are used in the decoder to provide the global feature map. With dynamic convolution, each instance can be segmented independently, reducing interference with other instances and improving segmentation accuracy. Under the supervision of projection loss and pixel point colour pairing loss, the contours of each object are finally outlined. On the PASCAL VOC and MS COCO datasets, the proposed method is competitive with more sophisticated approaches. In the VOC dataset, segmentation performance achieved 37.6% average precision with ResNet‐101 and FPN networks. The extensively visualised results demonstrate the effectiveness of the proposed encoder‐decoder framework with dynamic convolution.

Lower versions of EfficientDet (such as D0, D1) have smaller network structures and parameter sizes, but lower detection accuracy. Higher versions exhibit higher accuracy, but the increase in network complexity poses challenges for real-time processing and hardware requirements. To meet the higher accuracy requirements under limited computational resources, this paper introduces SpanEffiDet based on the channel adaptive frequency filter (CAFF) and the Span-Path Bidirectional Feature Pyramid structure. Firstly, the CAFF module proposed in this paper realizes the frequency domain transformation of channel information through Fourier transform and effectively extracts the key features through semantic adaptive frequency filtering, thus, eliminating channel redundant information of EfficientNet. Simultaneously, the module has the ability to compute the weights across the channels and at fine granularity, and capture the detailed information of element features. Secondly, a two-way characteristic pyramid network multi-level cross-BIFPN, which can achieve multi-layer and multi-nodes, is proposed to build cross-level information transmission to incorporate both semantic and positional information of the target. This design enables the network to more effectively detect objects with significant size differences in complex environments. Finally, by introducing generalized focal Loss V2, reliable localization quality estimation scores are predicted from the distribution statistics of bounding boxes, thereby improving localization accuracy. The experimental results indicate that on the MS COCO dataset, SpanEffiDet-D0 achieved an AP improvement of 3.3% compared to the original EfficientDet series algorithms. Similarly, on the PASCAL VOC2007 and 2012 datasets, the mAP of SpanEffiDet-D0 is respectively 1.66 and 2.65% higher than that of EfficientDet-D0.

Object detection, which is one of the most fundamental visual recognition tasks, has been a hotspot in computer vision. CNN (Convolutional Neural Networks) have been widely employed for building detector. Due to the success of RPN (Region Proposal Network), the two-stage detectors get both classification accuracy and precise regression bounding boxes. However, they still struggle in small-size object detection. In this paper, we present a deep network, namely Spatial Fine-Grained Network (SFGN). The SFGN that exploits Spatial Fine-Grained Features (SFGF) concatenates the higher resolution features, which is fine-grained with the low resolution features and high-level semantic by stacking spatial features for fine-grained features. An enhanced region proposal generator is proposed to get the objectless for small object to obtain a small set of proposal. The contextual information surrounding the region of interest is embedded using local spatial information for increasing the useful information and discriminating the background. For improving the detection performance, we use a simple yet surprisingly effective online hard example mining (OHEM) algorithm for training region proposal generator. It embeds an efficiently implemented soft non-maximum suppression (soft-NMS) for replacing with tradition NMS to obtain consistent improvements without increasing the computational complexity in inference. On PASCAL VOC 2007 and PASCAL VOC 2012 datasets, our SFGN improves baseline model from 81.2% mAP to 80.6% mAP. On MS COCO dataset, SFGN also performs better than baseline model. As intuition suggests, our detection results provide strong evidence that our SFGN improves detection accuracy, especially in small object test.

Objects for detection usually have distinct characteristics in different sub-regions and different aspect ratios. However, in prevalent two-stage object detection methods, Region-of-Interest (RoI) features are extracted by RoI pooling with little emphasis on these translation-variant feature components. We present feature selective networks to reform the feature representations of RoIs by exploiting their disparities among sub-regions and aspect ratios. Our network produces the sub-region attention bank and aspect ratio attention bank for the whole image. The RoI-based sub-region attention map and aspect ratio attention map are selectively pooled from the banks, and then used to refine the original RoI features for RoI classification. Equipped with a lightweight detection subnetwork, our network gets a consistent boost in detection performance based on general ConvNet backbones (ResNet-101, GoogLeNet and VGG-16). Without bells and whistles, our detectors equipped with ResNet-101 achieve more than 3% mAP improvement compared to counterparts on PASCAL VOC 2007, PASCAL VOC 2012 and MS COCO datasets.

Scale variation is one of the important challenges in object detection. Many state-of-the-art objectors tackle this problem by utilizing the feature pyramids. However, the current methods of producing feature pyramids are still inefficient to integrate the semantic information from other layers. In this work, our motivation is to build a feature pyramid efficiently with the selected contextual feature by integrating the informative features and suppressing the useless ones. To achieve this goal, we propose a novel single-stage detection network termed Selective Feature Network(SFNet) which consists of a semantic-enhanced module and a selective feature module. The semantic-enhanced module improves the semantics of basic pyramids via a lightweight architecture. In conjunction with that, a selective feature module is employed to combine features across different channels and scales by attention mechanism. The resulting contextual feature is then injected into the pyramidal features. Comprehensive experiments are performed on PASCAL VOC and MS COCO datasets. Results demonstrate that, with a VGG16 based SFNet, our approach obtains significant improvements over the competitors without losing real-time processing speed.

Feature pyramid network is widely used in advanced object detection. By simply changing the network connection, the performance of small object detection can be greatly improved without increasing the amount of calculation of the original model. However, the algorithm still has some shortcomings. Therefore, a new attention feature pyramid network (attention PFN) is proposed. Firstly, an improved receptive field module is added, which can make full use of global and local context information. Secondly, the connection mode of the pyramid is further optimized. Deconvolution is used to replace the nearest neighbor interpolation in top-down up-sampling and a channel attention module is added to the horizontal connection to highlight important context information. Finally, adaptive fusion and spatial feature balance are used for each feature pyramid, so that the network can learn the weights of different feature layers. Each pyramid layer contains more discrimination information. Attention PFN is tested on Pascal VOC and MS COCO datasets, respectively. The experiment results revealed that the proposed method has better performance than the original algorithm. Therefore, the attention PFN is an effective algorithm.

In this paper, we study object detection using a large pool of unlabeled images and only a few labeled images per category, named "few-example object detection". The key challenge consists in generating trustworthy training samples as many as possible from the pool. Using few training examples as seeds, our method iterates between model training and high-confidence sample selection. In training, easy samples are generated first and, then the poorly initialized model undergoes improvement. As the model becomes more discriminative, challenging but reliable samples are selected. After that, another round of model improvement takes place. To further improve the precision and recall of the generated training samples, we embed multiple detection models in our framework, which has proven to outperform the single model baseline and the model ensemble method. Experiments on PASCAL VOC'07, MS COCO'14, and ILSVRC'13 indicate that by using as few as three or four samples selected for each category, our method produces very competitive results when compared to the state-of-the-art weakly-supervised approaches using a large number of image-level labels.