Running deep visual analytics models for real-time applications is challenging for mobile devices. Offloading the computation to edge server can mitigate computation bottleneck at the mobile device, but may decrease the analytics performance due to the necessity of compressing the image data. We consider a “split computing” system to offload a part of the deep learning model's computation and introduce a novel learned feature compression approach with lightweight computation. We demonstrate the effectiveness of the split computing pipeline in performing computation offloading for the problems of object detection and image classification. Compared to compressing the raw images at the mobile, and running the analytics model on the decompressed images at the server, the proposed feature-compression approach can achieve significantly higher analytics performance at the same bit rate, while reducing the complexity at the mobile. We further propose a scalable feature compression approach, which facilitates adaptation to network bandwidth dynamics, while having comparable performance to the non-scalable approach.

Object detection is a fundamental task for autonomous driving, which aim to identify and localize objects within an image. Deep learning has made great strides for object detection, with popular models including Faster R-CNN, YOLO, and SSD. The detection accuracy and computational cost of object detection depend on the spatial resolution of an image, which may be constrained by both the camera and storage considerations. Furthermore, original images are often compressed and uploaded to a remote server for object detection. Compression is often achieved by reducing either spatial or amplitude resolution or, at times, both, both of which have well-known effects on performance. Detection accuracy also depends on the distance of the object of interest from the camera. Our work examines the impact of spatial and amplitude resolution, as well as object distance, on object detection accuracy and computational cost. As existing models are optimized for uncompressed (or lightly compressed) images over a narrow range of spatial resolution, we develop a resolution-adaptive variant of YOLOv5 (RA-YOLO), which varies the number of scales in the feature pyramid and detection head based on the spatial resolution of the input image. To train and evaluate this new method, we created a dataset of images with diverse spatial and amplitude resolutions by combining images from the TJU and Eurocity datasets and generating different resolutions by applying spatial resizing and compression. We first show that RA-YOLO achieves a good trade-off between detection accuracy and inference time over a large range of spatial resolutions. We then evaluate the impact of spatial and amplitude resolutions on object detection accuracy using the proposed RA-YOLO model. We demonstrate that the optimal spatial resolution that leads to the highest detection accuracy depends on the ‘tolerated’ image size (constrained by the available bandwidth or storage). We further assess the impact of the distance of an object to the camera on the detection accuracy and show that higher spatial resolution enables a greater detection range. These results provide important guidelines for choosing the image spatial resolution and compression settings predicated on available bandwidth, storage, desired inference time, and/or desired detection range, in practical applications.

Recent works on learned image coding using autoencoder models have achieved promising results in rate-distortion performance. Typically, an autoencoder is used to transform an image into a latent tensor, which is then quantized and entropy coded. Based on a work by Ballé et al., we adapted the autoencoder with a hyperprior model to code images in a block-based approach. When the autoencoder model is directly applied to code small image blocks, spatial redundancy in the larger image cannot be fully utilized, resulting in a decrease in ratedistortion performance. We propose a method to utilize border information in the entropy coding of latent and hyper-latent tensors, which has achieved promising results. We show that using intra-prediction to help entropy coding is more effective than applying a convolutional autoencoder with hyper priors to intra-prediction residual blocks.