PyTorch code for VSRN described in the paper "Visual Semantic Reasoning for Image-Text Matching". The paper will appear in ICCV 2019 as oral presentation. It is built on top of the VSE++.
Kunpeng Li, Yulun Zhang, Kai Li, Yuanyuan Li and Yun Fu. "Visual Semantic Reasoning for Image-Text Matching", ICCV, 2019. [pdf]
Image-text matching has been a hot research topic bridging the vision and language areas. It remains challenging because the current representation of image usually lacks global semantic concepts as in its corresponding text caption. To address this issue, we propose a simple and interpretable reasoning model to generate visual representation that captures key objects and semantic concepts of a scene. Specifically, we first build up connections between image regions and perform reasoning with Graph Convolutional Networks to generate features with semantic relationships. Then, we propose to use the gate and memory mechanism to perform global semantic reasoning on these relationship-enhanced features, select the discriminative information and gradually generate the representation for the whole scene.
Experiments validate that our method achieves a new state-of-the-art for the image-text matching on MS-COCO and Flickr30K datasets. It outperforms the current best method SCAN by 6.8% relatively for image retrieval and 4.8% relatively for caption retrieval on MS-COCO (Recall@1 using 1K test set). On Flickr30K, our model improves image retrieval by 12.6% relatively and caption retrieval by 5.8% relatively (Recall@1).
Besides, since our method only relies on the simple inner product as the similarity function, it is quite efficient at the inference stage. It is around 30 times faster than the current best method SCAN when tested on MS-COCO 1K dataset.
We recommended the following dependencies.
import nltk
nltk.download()
> d punkt
Download the dataset files and pre-trained models. We use splits produced by Andrej Karpathy.
We follow bottom-up attention model and SCAN to obtain image features for fair comparison. More details about data pre-processing (optional) can be found here. All the data needed for reproducing the experiments in the paper, including image features and vocabularies, can be downloaded from SCAN by using:
wget https://scanproject.blob.core.windows.net/scan-data/data.zip
You can also get the data from google drive: https://drive.google.com/drive/u/1/folders/1os1Kr7HeTbh8FajBNegW8rjJf6GIhFqC. We refer to the path of extracted files for data.zip
as $DATA_PATH
.
Modify the model_path and data_path in the evaluation_models.py file. Then Run evaluation_models.py
:
python evaluation_models.py
To do cross-validation on MSCOCO 1K test set (5 folders average), pass fold5=True
. Pass fold5=False
for evaluation on MSCOCO 5K test set. Pretrained models for MSCOCO and Flickr30K can be downloaded from https://drive.google.com/file/d/1y8Ywa2vrPB7m_Q_Ku69z7EdwsLB9gsJW/view?usp=sharing
You can also use the following code to evaluate each single model on Flickr30K, MSCOCO 1K and MSCOCO 5K :
from vocab import Vocabulary
import evaluation
evaluation.evalrank("pretrain_model/flickr/model_fliker_1.pth.tar", data_path="$DATA_PATH", split="test", fold5=False)'
evaluation.evalrank("pretrain_model/coco/model_coco_1.pth.tar", data_path="$DATA_PATH", split="testall", fold5=True)'
evaluation.evalrank("pretrain_model/coco/model_coco_1.pth.tar", data_path="$DATA_PATH", split="testall", fold5=False)'
Run train.py
:
For MSCOCO:
python train.py --data_path $DATA_PATH --data_name coco_precomp --logger_name runs/coco_VSRN --max_violation
For Flickr30K:
python train.py --data_path $DATA_PATH --data_name f30k_precomp --logger_name runs/flickr_VSRN --max_violation --lr_update 10 --max_len 60
If you found this code useful, please cite the following paper:
@inproceedings{li2019vsrn,
title={Visual semantic reasoning for image-text matching},
author={Li, Kunpeng and Zhang, Yulun and Li, Kai and Li, Yuanyuan and Fu, Yun},
booktitle={ICCV},
year={2019}
}