Split-computing has recently emerged as a paradigm for offloading computation of visual analytics models from low-powered mobile devices to edge or cloud servers, by which the mobiles execute part of the model and compress and send the intermediate features, and the servers complete the remaining model computation. Prior feature compression approaches train different compression models and possibly visual analytics models to reach different target bit rates. We propose a scalable compression model that compresses the intermediate features of the YOLO object detection model into a layered bitstream, which can be easily adapted to meet the rate constraint of a dynamic network. Our approach achieves comparable rate-accuracy performance compared to prior non-scalable compression approaches over a large bitrate range.
more »
« less
Feature Compression for Rate Constrained Object Detection on the Edge
Recent advances in computer vision has led to a growth of interest in deploying visual analytics model on mobile devices. However, most mobile devices have limited computing power, which prohibits them from running large scale visual analytics neural networks. An emerging approach to solve this problem is to offload the computation of these neural networks to computing resources at an edge
server. Efficient computation offloading requires optimizing the trade-off between multiple objectives including compressed data rate, analytics performance, and computation
speed. In this work, we consider a “split computation” system to offload a part of the computation of the YOLO object detection model. We propose a learnable feature compression approach to compress the intermediate YOLO features with lightweight computation. We train the feature compression and decompression module together with the YOLO model to optimize the object detection accuracy
under a rate constraint. Compared to baseline methods that apply either standard image compression or learned image compression at the mobile and perform image de-compression and YOLO at the edge, the proposed system achieves higher detection accuracy at the low to medium rate range. Furthermore, the proposed system requires substantially lower computation time on the mobile device with CPU only.
more »
« less
- Award ID(s):
- 1952180
- NSF-PAR ID:
- 10466150
- Date Published:
- Journal Name:
- In MLSys 2023 Workshop on Resource-Constrained Learning in Wireless Networks
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Efficient and adaptive computer vision systems have been proposed to make computer vision tasks, such as image classification and object detection, optimized for embedded or mobile devices. These solutions, quite recent in their origin, focus on optimizing the model (a deep neural network, DNN) or the system by designing an adaptive system with approximation knobs. Despite several recent efforts, we show that existing solutions suffer from two major drawbacks. First , while mobile devices or systems-on-chips (SOCs) usually come with limited resources including battery power, most systems do not consider the energy consumption of the models during inference. Second , they do not consider the interplay between the three metrics of interest in their configurations, namely, latency, accuracy, and energy. In this work, we propose an efficient and adaptive video object detection system — Virtuoso , which is jointly optimized for accuracy, energy efficiency, and latency. Underlying Virtuoso is a multi-branch execution kernel that is capable of running at different operating points in the accuracy-energy-latency axes, and a lightweight runtime scheduler to select the best fit execution branch to satisfy the user requirement. We position this work as a first step in understanding the suitability of various object detection kernels on embedded boards in the accuracy-latency-energy axes, opening the door for further development in solutions customized to embedded systems and for benchmarking such solutions. Virtuoso is able to achieve up to 286 FPS on the NVIDIA Jetson AGX Xavier board, which is up to 45 times faster than the baseline EfficientDet D3 and 15 times faster than the baseline EfficientDet D0. In addition, we also observe up to 97.2% energy reduction using Virtuoso compared to the baseline YOLO (v3) — a widely used object detector designed for mobiles. To fairly compare with Virtuoso , we benchmark 15 state-of-the-art or widely used protocols, including Faster R-CNN (FRCNN) [NeurIPS’15], YOLO v3 [CVPR’16], SSD [ECCV’16], EfficientDet [CVPR’20], SELSA [ICCV’19], MEGA [CVPR’20], REPP [IROS’20], FastAdapt [EMDL’21], and our in-house adaptive variants of FRCNN+, YOLO+, SSD+, and EfficientDet+ (our variants have enhanced efficiency for mobiles). With this comprehensive benchmark, Virtuoso has shown superiority to all the above protocols, leading the accuracy frontier at every efficiency level on NVIDIA Jetson mobile GPUs. Specifically, Virtuoso has achieved an accuracy of 63.9%, which is more than 10% higher than some of the popular object detection models, FRCNN at 51.1%, and YOLO at 49.5%.more » « less
-
Localization in urban environments is becoming increasingly important and used in tools such as ARCore [ 18 ], ARKit [ 34 ] and others. One popular mechanism to achieve accurate indoor localization and a map of the space is using Visual Simultaneous Localization and Mapping (Visual-SLAM). However, Visual-SLAM is known to be resource-intensive in memory and processing time. Furthermore, some of the operations grow in complexity over time, making it challenging to run on mobile devices continuously. Edge computing provides additional compute and memory resources to mobile devices to allow offloading tasks without the large latencies seen when offloading to the cloud. In this article, we present Edge-SLAM, a system that uses edge computing resources to offload parts of Visual-SLAM. We use ORB-SLAM2 [ 50 ] as a prototypical Visual-SLAM system and modify it to a split architecture between the edge and the mobile device. We keep the tracking computation on the mobile device and move the rest of the computation, i.e., local mapping and loop closing, to the edge. We describe the design choices in this effort and implement them in our prototype. Our results show that our split architecture can allow the functioning of the Visual-SLAM system long-term with limited resources without affecting the accuracy of operation. It also keeps the computation and memory cost on the mobile device constant, which would allow for the deployment of other end applications that use Visual-SLAM. We perform a detailed performance and resources use (CPU, memory, network, and power) analysis to fully understand the effect of our proposed split architecture.more » « less
-
Localization in urban environments is becoming increasingly important and used in tools such as ARCore [11], ARKit [27] and others. One popular mechanism to achieve accurate indoor localization as well as a map of the space is using Visual Simultaneous Localization and Mapping (Visual-SLAM). However, Visual-SLAM is known to be resource-intensive in memory and processing time. Further, some of the operations grow in complexity over time, making it challenging to run on mobile devices continuously. Edge computing provides additional compute and memory resources to mobile devices to allow offloading of some tasks without the large latencies seen when offloading to the cloud. In this paper, we present Edge-SLAM, a system that uses edge computing resources to offload parts of Visual-SLAM. We use ORB-SLAM2 as a prototypical Visual-SLAM system and modify it to a split architecture between the edge and the mobile device. We keep the tracking computation on the mobile device and move the rest of the computation, i.e., local mapping and loop closure, to the edge. We describe the design choices in this effort and implement them in our prototype. Our results show that our split architecture can allow the functioning of the Visual-SLAM system long-term with limited resources without affecting the accuracy of operation. It also keeps the computation and memory cost on the mobile device constant which would allow for deployment of other end applications that use Visual-SLAM.more » « less
-
more » « less
Network pruning is a widely used technique to reduce computation cost and model size for deep neural networks. However, the typical three-stage pipeline (i.e., training, pruning, and retraining (fine-tuning)) significantly increases the overall training time. In this article, we develop a systematic weight-pruning optimization approach based on surrogate Lagrangian relaxation (SLR), which is tailored to overcome difficulties caused by the discrete nature of the weight-pruning problem. We further prove that our method ensures fast convergence of the model compression problem, and the convergence of the SLR is accelerated by using quadratic penalties. Model parameters obtained by SLR during the training phase are much closer to their optimal values as compared to those obtained by other state-of-the-art methods. We evaluate our method on image classification tasks using CIFAR-10 and ImageNet with state-of-the-art multi-layer perceptron based networks such as MLP-Mixer; attention-based networks such as Swin Transformer; and convolutional neural network based models such as VGG-16, ResNet-18, ResNet-50, ResNet-110, and MobileNetV2. We also evaluate object detection and segmentation tasks on COCO, the KITTI benchmark, and the TuSimple lane detection dataset using a variety of models. Experimental results demonstrate that our SLR-based weight-pruning optimization approach achieves a higher compression rate than state-of-the-art methods under the same accuracy requirement and also can achieve higher accuracy under the same compression rate requirement. Under classification tasks, our SLR approach converges to the desired accuracy × faster on both of the datasets. Under object detection and segmentation tasks, SLR also converges 2× faster to the desired accuracy. Further, our SLR achieves high model accuracy even at the hardpruning stage without retraining, which reduces the traditional three-stage pruning into a two-stage process. Given a limited budget of retraining epochs, our approach quickly recovers the model’s accuracy.
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
