If you use --recursive
to clone this repository, skip this section.
Please run following script to download evaluation codes for Microsoft COCO Caption Evaluation and put them under the coco-caption
directory:
cd GL-RG
git clone https://github.com/tylin/coco-caption.git
Please refer to Microsoft COCO Caption Evaluation for more setup details.
Please run following script to download Consensus-based Image Description Evaluation and put them under the cider
directory:
cd GL-RG
git clone https://github.com/plsang/cider.git
If you have cloned this repository from this GitHub, skip this section.
Download the metadate
and feature
from Here and then unzip them under the data
directory like:
data
├── feature
│ ├── msrvtt_test_sem_tag_res.h5
│ ├── msrvtt_train_sem_tag_res.h5
│ ├── msrvtt_val_sem_tag_res.h5
│ ├── msvd_test_sem_tag_res.h5
│ ├── msvd_train_sem_tag_res.h5
│ └── msvd_val_sem_tag_res.h5
│
├── metadata
│ ├── msrvtt_test_cocofmt.json
│ ├── msrvtt_test_sequencelabel.h5
│ ├── msvd_test_cocofmt.json
└── └── msvd_test_sequencelabel.h5
Download the pre-trained weights from Here and then unzip them under the model
directory like:
model
├── GL-RG_XE_msrvtt
│ └── model.pth
│
├── GL-RG_XE_msvd
│ └── model.pth
│
├── GL-RG_DXE_msrvtt
│ └── model.pth
│
├── GL-RG_DXE_msvd
│ └── model.pth
│
├── GL-RG_DR_msrvtt
│ └── model.pth
│
└── GL-RG_DR_msvd
└── model.pth