hucvl.github.io

Paper

Kemal CIZMECILER, Erkut Erdem and Aykut Erdem. “Leveraging Semantic Saliency Maps for Query-Specific Video Summarization “, has been accepted for publication in Multimedia Tools and Applications. pdf

ARS DATASET

This dataset contains video sets collected from publicly available Youtube videos.

we focus on crowd videos as they can be annotated with a rich set of attributes about the observed activities, actors and events. Here we use complex query terms such as ‘fight between fans’ and ‘running and swimming’. In total, we collected 10 videos, including music and movie clips with 720 × 1280 resolution. The duration of the videos changes between 2-12 minutes. More specifically, there is a video log about a triathlon race, which contains running, swimming and biking activities. There exist 3 music videos, mainly showing cheering fans in stadiums or streets. There are 3 movie clips, which respectively show a battlefield scene with a ceremony in a church, dancers in a car and in a club, and finally an orchestra band on a stage with full of dancing people in front of them. Lastly, there is a rock festival video, which contains singing, resting on a beach and playing activities.

Video ID Video name Attribute Count Number of Frames Duration Number of Queries Shot Count Youtube Link
1 Triathlon 19 10971 12:12 15 82 k17Kg1owKSw
2 LetItBe 20 5623 03:54 23 75 ShNnEDb4wFA
3 HoldMeTight 13 3393 01:53 12 37 JdzSuH-azJs
4 FansFighting 17 14832 08:14 13 48 hjowpb2maAM
5 WavinFlag 21 6400 04:16 7 45 TR2o9GY8u60
6 KNAANMusicVideo 22 6736 03:44 11 70 WTJSt4wP2ME
7 TarkanMusicClip 17 6876 04:35 13 81 EBwjmeDoE6A
8 ThePianist 18 10831 06:01 16 59 DENCBZF0wd8
9 RockFestival 16 10734 05:57 18 42 hRgcDHO2eII
10 ComeTogether 14 6819 03:48 14 38 wH6eiCJ9SDs
  Total 60(unique) 91339 54:36 142 577  

How To Download

Attribute list and annotations are provided as as mat file, and ground truth human summaries also provided as .txt files. usersummaries_annotations.zip

Queries in videos are here attributes_in_queries.zip

Example comparison from ARS dataset can be seen as

ARS_example2

Example comparison from RAD dataset can be seen as

RAD_example2

Video List Used in the Paper from RAD Dataset

Sequence Title Query Duration Number of Frames Shot Count
5slean lean 01:04 1536 64
basketball fights fight 03:03 4368 182
bts dance girl group dance 02:51 4104 171
bts dance cover dance 02:48 4032 168
bus crash crash 02:00 2880 120
bus driver beats up girl beat 01:34 2256 94
chase bryant chase 03:14 4656 194
chase pig commercial chase 01:46 2544 106
chase utley slide chase 01:05 1560 65
cooking channel cook 02:13 3192 133
cooking videos cook 02:48 4032 168
cooking with dogs cook 03:00 4320 180
cooking with kylie cook 02:12 3168 132
dance practice dope dance 02:01 2904 121

Acknowledgements

This work was supported in part by GEBIP 2018 Award of the Turkish Academy of Sciences to E. Erdem, BAGEP 2021 Award of the Science Academy to A. Erdem.