Automatic video caption generation with STA-FG framework
This code is used for the implementation of models which described in the paper titled Fused-GRU for with Semantic-Temporal Attention for Video Captioning
- Python 2.7
The simplest way to install it is to useAnnaconda
. - Theano
We recomend you to use the newest version. Simply typepip install --upgrade https://github.com/Theano/Theano/archive/master.zip
to install the newest version. - Lasagne
Our implementation is based on the lasagne. Typepip install --upgrade https://github.com/Lasagne/Lasagne/archive/master.zip
to install the newest version of Lasagne. - coco-caption
Dowload this evaluation code. You need to add the path into your #PYTHONPATH or move it to the ${ROOTPATH}.
To run this code, please follow these instructions.
- You need to create a dataset folder and dowload preproccessed dataset here. Then unzip the msvd.zip file which in the "data" folder into the dataset folder.
- Revise config file: tools/cfg.py and set your dataset path.
- Revise data processing file: tools/data_reader.py and reset your features path.
- run
scripts/script_train.py
to train a model and test it. - run
scripts/script_eval_stafg_rc.py
to evaluate models under the mutiple features setting.