Indexed by:
Abstract:
Live video platforms have attracted many active streamers and daily users, and quickly understanding live video streaming scenes is crucial for ensuring the clean and healthy cyberspace. Video logo often appear in live video and can serve as key clues for understanding video streaming scenes. Due to the presence of the jitter scenes, as well as blur and deformed logos in live streaming scenes, we propose a spatial-channel specific snake-yolov8 for video logo detection in live streaming scenes. First, we design an involutional fusion layer to fuse interframe logo features for jitter scenes. Then, the spatial-channel specific involution is proposed to extract spatial and channel context, to adapt to blurring logos during logo detection. Finally, to cope with deformable logos in live steaming scenes, we embed snake deformation convolution into spatial-channel specific involution. Experimental results show that under inference efficiency of 39.8 FPS, the mAP of the proposed method reaches 69.7% on LogoDet-3K, and 53.0% on self-built BJUT-VLD, which has certain effectiveness and superiority for video logo detection in live streaming scenes. © 2024 ACM.
Keyword:
Reprint Author's Address:
Email:
Source :
Year: 2024
Page: 402-408
Language: English
Cited Count:
SCOPUS Cited Count:
ESI Highly Cited Papers on the List: 0 Unfold All
WanFang Cited Count:
Chinese Cited Count:
30 Days PV: 0
Affiliated Colleges: