Outstanding Paper Award at ICONI 2020
Files
The 12th International Conference on Internet
Outstanding Paper Award
Choi Ki-seok, a doctoral student in the University's Department of Computer Science & Engineering, and Professor Nang Jong-ho of the same department (corresponding author) received an Outstanding Paper Award at the ICONI 2020 conference, held from 13 to 16 December, for 'CNN-based Place Recognition with Attention Method in Broadcasting Video'.
△ From left: doctoral student Choi Ki-seok and Professor Nang Jong-ho (corresponding author)
In the paper the authors designed and proposed a place recognition model based on multiple image inputs in order to recognize places in broadcast content video. They tested the performance of various models according to whether multiple image sampling methods, attention over images, and fusion methods such as NetVLAD were applied. For the tests they defined 47 kinds of place and built a dataset by collecting 8 films, 10 dramas and 24 sitcoms.
The research confirmed that even when an image set includes images unsuited to background recognition, adjusting weights with an attention method yields higher performance than conventional single-image background recognition. The proposed place recognition model achieved 0.35 (top-1) and 0.66 (top-5) on the dataset in terms of mAP.
Choi Ki-seok is due to receive his degree this year and has founded a start-up. He said: "I am currently pursuing business related to automated video content analysis."
△ ICONI 2020 Outstanding Paper Award