Wireless Image Retrieval at the Edge

We study the image retrieval problem at wireless edge, where an edge device captures image, which is then used to retrieve similar images from server. These can be of same person or a vehicle taken other cameras different times and locations. Our goal maximize accuracy task under power bandwidth constraints over link. Due stringent delay constraint underlying application, sending whole sufficient quality not possible. propose two alternative schemes based on digital analog communications, respectively. In approach, we first deep neural network (DNN) aided retrieval-oriented compression scheme, whose output bit sequence transmitted channel using conventional codes. joint source coding (JSCC) feature vectors are directly mapped into symbols. evaluate both re-identification (re-ID) tasks conditions, including static fading channels. show that JSCC scheme significantly increases end-to-end accuracy, speeds up encoding process, provides graceful degradation with conditions. The proposed architecture evaluated through extensive simulations datasets as well ablation studies.