![]() ![]() SPPĮxtrapolating benchmarking results from Table 2 and Table 3, it can be observed that the proposed object detector, NextDet, achieves a notable performance in Argoverse’s monocular streaming perception as well as general-purpose multiclass object detection in Microsoft COCO datasets. Figure 6 provides a simple graphical representation of this technique. The bottom-up approach provides strong positioning features, and the top-down approach provides strong semantic features, resulting in an improvement in object detection performance. To address this issue, the PAN is connected to the FPN and up sampled by a factor of two using the nearest neighbor approach, allowing bottom features to be connected to the top features. However, since the feature maps are of different sizes in the feature pyramid, the bottom features cannot be fused with the features on the top. The FPN provides feature maps of different sizes in order to fuse different features together. The neck of the proposed object detector is implemented by connecting the PAN to the FPN. The neck module of a modern object detector is linked to the backbone module which acts like a feature aggregator by collecting feature maps from different stages of the backbone and fusing them together with the help of pyramid networks such as Feature Pyramid Networks (FPN) and Path Aggregation Network (PANet or PAN). Figure 2 provides a visual comparison between a one-stage object detector and a two-stage object detector. State-of-the-art YOLO algorithm is a popular choice in designing modern object detectors because of its speed and accuracy performance. In comparison, YOLO obtains a better inference performance at the cost of detection performance whereas SSD obtains a better detection performance at the cost of inference performance. On the other hand, one-stage detectors such as YOLO and SSD execute image classification and object detection tasks in a single module but may not be able to achieve a desirable real-time inference performance due to constrained computational resources on edge devices. Two-stage detectors such as the family of R-CNN perform these tasks separately in different modules by utilizing a Region Proposal Network in the first stage to generate sparse region proposals to obtain Region of Interest (RoI) and then passing down these region proposals for object classification and bounding-box regressions in the second stage resulting in an increase in object detection accuracy but leading to a time complexity bottleneck and therefore, an increase in demand of computational resources for implementation purposes. Extensive experiments and ablation tests, as outlined in this paper, are performed on Argoverse-HD and COCO datasets, which provide numerous temporarily sparse to dense annotated images, demonstrate that the proposed object detection algorithm with CondenseNeXt as the backbone result in an increase in mean Average Precision (mAP) performance and interpretability on Argoverse-HD’s monocular ego-vehicle camera captured scenarios by up to 17.39% as well as COCO’s large set of images of everyday scenes of real-world common objects by up to 14.62%. The scope of the work presented within this paper proposes a modern object detection network called NextDet to efficiently detect objects of multiple classes which utilizes CondenseNeXt, an award-winning lightweight image classification convolutional neural network algorithm with reduced number of FLOPs and parameters as the backbone, to efficiently extract and aggregate image features at different granularities in addition to other novel and modified strategies such as attentive feature aggregation in the head, to perform object detection and draw bounding boxes around the detected objects. Object detection is a computer vision task of detecting instances of objects of a certain class, identifying types of objects, determining its location, and accurately labelling them in an input image or a video.
0 Comments
Leave a Reply. |