We released the dataset for venue category estimation from micro-videos.
Micro-videos spread rapidly across various onlineflagship platforms, such as Instagram,Snapchat, and Vine, since the late of 2012. We aim to label such bite-sized video clips with venue categories.In this repository, we relased a rich set of feature extracted from micro-videos which crawled from Vine. In particular,our dataset is consisting of 270,145 micro-videos distributed in 188 Foursquare venue categories (VENUE-188 for short). We further splited VENUE-188 into training,valid,and testing data in a ratio of 50%, 20%,and 30%, with a number of 132370,56730,and 81044,respectively. Besides,the corresponding videos-ids were also recorded in the dataset.
There are several research tasks can be conducted in the VENUE-188.
-
multi-modal venue category estimation
-
mono-modal venue categry estimation
-
The labels (from 1-188) and their corresponding venue cateogries classes;
-
alex_net: alexnet_visual_feature (4096 dim) + stacked_denosing_autoencode_feature(200 dim) + paragraph_textual_feature(100 dim). You can access this feature set via this link:.
-
inception_v3: inceptionV3_visual_feature (2048 dim) + stacked_denosing_autoencode_feature(200 dim) + paragraph_textual_feature(100 dim). You can access this feature set via this link:https://pan.baidu.com/s/1c2vh1DI.
-
vgg19: vgg19_visual_feature (512 dim)+ stacked_denosing_autoencode_feature(200 dim) + paragraph_textual_feature(100 dim).You can access this feature set via this link:https://pan.baidu.com/s/1kULbYMr
-
resnet50: resnet50_visual_feature (2048 dim) + stacked_denosing_autoencode_feature(200 dim) + paragraph_textual_feature(100 dim). You can access this feature via this link:https://pan.baidu.com/s/1mhS0Pp2.
-
video description:user generated text and hashtags:https://pan.baidu.com/s/1bpnbWAz.
-
video url:the url of all downloaded videos.https://pan.baidu.com/s/1mioMh4W.
Please cite it as...
@inproceedings{Zhang2016Shorter,
title={Shorter-is-Better: Venue Category Estimation from Micro-Video},
author={Zhang, Jianglong and Nie, Liqiang and Wang, Xiang and He, Xiangnan and Huang, Xianglin and Chua, Tat Seng},
booktitle={ACM on Multimedia Conference},
pages={1415-1424},
year={2016},
}
- All code in this repository is under the MIT license as specified by the LICENSE file.
- The ResNet50 weights are ported from the ones released by Kaiming He under the MIT license.
- The VGG16 and VGG19 weights are ported from the ones released by VGG at Oxford under the Creative Commons Attribution License.