The CASE Dataset of Candidate Spaces for Advert Implantation

Abstract

With the advent of faster internet services and growth of multimedia content, we observe a massive growth in the number of online videos. The users generate these video contents at an unprecedented rate, owing to the use of smart-phones and other hand-held video capturing devices. This creates immense potential for the advertising and marketing agencies to create personalized content for the users. In this paper, we attempt to assist the video editors to generate augmented video content, by proposing candidate spaces in video frames. We propose and release a large-scale dataset of outdoor scenes, along with manually annotated maps for candidate spaces. We also benchmark several deep-learning based semantic segmentation algorithms on this proposed dataset.

SSD vs. YOLO for Detection of Outdoor Urban Advertising Panels under Multiple Variabilities

“…After that, we show the detection results for SSD and YOLOv3 under the different types of variabilities and compare them. In order to compare our results with those presented in the work by Dev et al [ 21 ], we introduce and apply the same additional measures used by these authors. Finally, we present a global discussion on results achieved in the work.…”

“…In order to compare our approach with the results presented by Dev et al [ 21 ], we have included some additional average performance semantic segmentation metrics to evaluate the accuracy of detections for SSD and YOLOv3 networks. The metrics are related with pixel classification accuracy and IoU, and they are Pixel Accuracy of Class i (PA i ), Mean Accuracy (MA), Mean IoU (M_IoU) and Frequency Weighted IoU (FW_IoU).…”

“…We have not found any publicly available dataset of outdoor urban panel images with the characteristics we are considering for our study (i.e., the corresponding ones to OPPI panels). A related referenced dataset of billboard adverts is CASE (CAndidate Spaces for advErt implantation) [ 21 ] which was built from the Cityscapes dataset [ 38 ] and includes street view images. The CASE dataset was created by randomly selecting 10,000 images from Cityscapes dataset, and annotating them with the placements of advertisements.…”

“…These same authors in 2019 have also published a related work [ 21 ] for automatically detecting existing billboards in videos and replacing the advertisements contained in them with new ones. The interest is focused in learning candidate placement of billboards in outdoor scenes in order to place regularly shaped billboards in street view images.…”

“…Previous works on billboard detection [ 18 , 21 ] have considered the detection problem as a semantic segmentation one, where classification and localization was performed at the level of image pixels. Moreover, the authors have used specific deep learning networks for such a semantic segmentation task.…”