Repository files navigation

This is the implementation for ICCV 17 paper "Temporal Context Network for Activity Localization in Videos".
If you use the code, pretrained models, proposals, please cite:

@InProceedings{Dai_2017_ICCV,
author = {Dai, Xiyang and Singh, Bharat and Zhang, Guyue and Davis, Larry S. and Qiu Chen, Yan},
title = {Temporal Context Network for Activity Localization in Videos},
booktitle = {The IEEE International Conference on Computer Vision (ICCV)},
month = {Oct},
year = {2017}
}

We provide the pre-trained proposal for both ActivityNet and THUMOS to assist future temporal detection works.

Prerequisite: A caffe with python support

Set PYTHONPATH to pycaffe path
Set ACTNET_HOME to folder with features

run "all_in_one.sh" to train and test

We fine-tuneTSN on dataset and extract score features.

For global features such as mbh and imagenet_shuffle, you can download from the official website.

Here are the pre-trained models: