Object detection is a hot topic with various applications in computer vision, e.g., image understanding, autonomous driving, and video surveillance. Much of the progresses have been driven by the availability of object detection benchmark datasets, including PASCAL VOC, ImageNet, and MS COCO. However, object detection on the drone platform is still a challenging task, due to various factors such as view point change, occlusion, and scales. To narrow the gap between current object detection performance and the real-world requirements, we organized the Vision Meets Drone (VisDrone2018) Object Detection in Image challenge in conjunction with the 15th European Conference on Computer Vision (ECCV 2018). Specifically, we release a large-scale drone-based dataset, including 8, 599 images (6, 471 for training, 548 for validation, and 1, 580 for testing) with rich annotations, including object bounding boxes, object categories, occlusion, truncation ratios, etc. Featuring a diverse real-world scenarios, the dataset was collected using various drone models, in different scenarios (across 14 different cities spanned over thousands of kilometres), and under various weather and lighting conditions. We mainly focus on ten object categories in object detection, i.e., pedestrian, person, car, van, bus, truck, motor, bicycle, awning-tricycle, and tricycle. Some rarely occurring special vehicles (e.g., machineshop truck, forklift truck, and tanker) are ignored in evaluation. The dataset is extremely challenging due to various factors, including large scale and pose variations, occlusion, and clutter background. We present the evaluation protocol of the VisDrone-DET2018 challenge and the comparison results of 38 detectors on the released dataset, which are publicly available on the challenge website: http://www.aiskyeye.com/ . We expect the challenge to largely boost the research and development in object detection in images on drone platforms.

Instance segmentation needs to locate all instances in an image correctly and segment each instance precisely. Currently, the most dominant methods for instance segmentation take object detection as a pre-task. However, they rely on the accuracy of object detection incredibly. If the pre-task cannot predict an accurate bounding box, the performance of instance segmentation will degenerate. In this paper, we present a novel method for instance segmentation to solve this problem, which is called S egmenting B eyond the B ounding B ox ( S3B-Net ). Our S3B-Net designs a sub-network to help instance segmentation methods based on object detection to segment the part of an instance beyond the bounding box. Specifically, the sub-network first predicts a two-dimensional pixel embedding for each pixel. Then, the Gaussian function is employed to calculate a pixel’s probability belongs to a corresponding instance according to the two-dimensional pixel embedding. Finally, the output of the sub-network combines with the output of instance segmentation based on object detection to generate a more precise instance mask. Our sub-network can easily extend on the existing instance segmentation method based on object detection to segment instance beyond the bounding box. We do our experiments on dominant instance segmentation datasets, such as the COCO dataset and Cityscapes dataset. The results show that our method can achieve 6.8 points gain compared with the baseline Mask R-CNN with ResNet-50-FPN in Cityscapes datasets, and 1.7 points gain with ResNet-101-FPN-DCN in COCO datasets. Our S3B-Net outperforms the previous state-of-the-art instance segmentation method, which proves our method is competitive. The source code of our method will be made available.

In object detection, precise object representation is a key factor to successfully classify and locate objects of an image. Existing methods usually use rectangular anchor boxes or a set of points to represent objects. However, these methods either introduce background noise or miss the continuous appearance information inside the object, and thus cause incorrect detection results. In this paper, we propose a novel anchor-free object detection network, called CrossDet++, which uses a set of growing crosslines along horizontal and vertical axes as object representations. An object can be flexibly represented as crosslines in different combinations, which inspires us to select the expressive crossline to effectively reduce the interference of noise. Meanwhile, the crossline representation takes into account the continuous adjacent object information, which is useful to enhance the discriminability of object features and find the object boundaries. Based on the learned crosslines, we propose an axis-query crossline growing module to adaptively capture features of crosslines and query surrounding pixels related to the line features for subsequent growing of crosslines. Their growing offsets and scales can be supervised by a decoupled regression mechanism, which limits the regression target to a specific direction for decreasing the optimization difficulty. During the training, we design a semantic-guided label assignment to emphasize the importance of crossline targets with higher semantic richness, further improving the detection performance. The experiment results demonstrate the effectiveness of our proposed method. Code can be available at: https://github.com/QiuHeqian/CrossDet.

Scene Parsing is a challenging task in computer vision, which can be formulated as a pixel-wise classification problem. Existing deep-learning-based methods usually use one general classifier to recognize all object categories. However, the general classifier easily makes some mistakes in dealing with some confusing categories that share similar appearances or semantics. In this paper, we propose an integrated classification model and a variance-based regularization to achieve more accurate classifications. On the one hand, the integrated classification model contains multiple classifiers, not only the general classifier but also a refinement classifier to distinguish the confusing categories. On the other hand, the variance-based regularization differentiates the scores of all categories as large as possible to reduce misclassifications. Specifically, the integrated classification model includes three steps. The first is to extract the features of each pixel. Based on the features, the second step is to classify each pixel across all categories to generate a preliminary classification result. In the third step, we leverage a refinement classifier to refine the classification result, focusing on differentiating the high-preliminary-score categories. An integrated loss with the variance-based regularization is used to train the model. Extensive experiments on three common scene parsing datasets demonstrate the effectiveness of the proposed method.

Class-incremental Semantic Segmentation (CISS) aims to learn new tasks sequentially that assign a specific category to each pixel of a given image while preserving the original capability to segment the old classes even if the labels of old tasks are absent. Most existing CISS methods suppress catastrophic forgetting by directly distilling on specific layers, which ignores the semantic gap between training data from the old and new classes with different distributions and leads to distillation errors, thus affecting segmentation performance. In this paper, we propose a Class-prompting Transformer (CPT) to introduce external prior knowledge provided by a pre-trained vision-language encoder into CISS pipelines for bridging the old and new classes and performing more generalized initialization and distillation. Specifically, we proposed a Prompt-guided Initialization Module (PIM), which measures the relationships between the class prompts and old query parameters to initialize the new query parameters for relocating the previous knowledge to the learning of new tasks. Then, a Semantic-aligned Distillation Module (SDM) is proposed to incorporate class prompt information with the class-aware embeddings extracted from the decoder to prevent the semantic gap problem between distinct class data and conduct adaptive knowledge transfer to suppress catastrophic forgetting. Extensive experiments on Pascal VOC and ADE20K datasets for the CISS task demonstrate the superiority of the proposed method, which achieves state-of-the-art performance.