Code based on our WACV 2022 Accepted Paper: https://arxiv.org/pdf/2110.02623.pdf

Project is built on top of the [CVSE] (https://github.com/BruceW91/CVSE) in PyTorch. However, it is easy to adapt to different Image-Text Matching models (SCAN, VSRN, SGRAF). Regarding the proposed metric code and evaluation, please visit: https://github.com/furkanbiten/ncs_metric.

The task of image-text matching aims to map representations from different modalities into a common joint visual-textual embedding. However, the most widely used datasets for this task, MSCOCO and Flickr30K, are actually image captioning datasets that offer a very limited set of relationships between images and sentences in their ground-truth annotations. This limited ground truth information forces us to use evaluation metrics based on binary relevance: given a sentence query we consider only one image as relevant. However, many other relevant images or captions may be present in the dataset. In this work, we propose two metrics that evaluate the degree of semantic relevance of retrieved items, independently of their annotated binary relevance. Additionally, we incorporate a novel strategy that uses an image captioning metric, CIDEr, to define a Semantic Adaptive Margin (SAM) to be optimized in a standard triplet loss. By incorporating our formulation to existing models, a large improvement is obtained in scenarios where available training data is limited. We also demonstrate that the performance on the annotated image-caption pairs is maintained while improving on other non-annotated relevant items when employing the full training set. The code for our new metric can be found at https://github.com/furkanbiten/ncs_metric and model https://github.com/andrespmd/semantic_adaptive_margin

Git clone the project.

Create Conda environment:

$ conda env create -f env.yml

Activate the environment:

$ conda activate pytorch12

Please download the following compressed file from:

https://drive.google.com/file/d/1DiRZODZQENoxbCpQFsiSEs1-9VatD176/view?usp=sharing

Uncompress the downloaded files under the main project folder.

Download the dataset files (MS-COCO and Flickr30K). We use regions of visual features, similar as the one used by SCAN, downloaded here. All the data needed for reproducing the experiments in the paper, including image features and vocabularies, can be downloaded from:

wget https://iudata.blob.core.windows.net/scan/data.zip
wget https://iudata.blob.core.windows.net/scan/vocab.zip

In this implementation, we refer to the path of extracted files for data.zip as $data_path and files for vocab.zip to ./vocab_path directory.

• Train MSCOCO models: Run train_coco.py using the SAM, e.a. using random sampling, metric margin division 6, weight on the cider loss 10 and on the 10% of the dataset:

python train_coco.py --data_path "$DATA_PATH" --model_name "$PATH_TO_SAVE" --resume None --val_step 1000 --use_metric --cider_weight 10 --metric_samples random --metric_div 6 --data_percentage 10

• Train Flickr30K models: Run train_f30k.py using the SAM, e.a. using soft sampling, metric margin division 10, weight on the cider loss 5 and on full dataset:

python train_f30k.py --data_path "$DATA_PATH" --model_name "$PATH_TO_SAVE" --resume None --val_step 1000 --use_metric --metric_samples soft --metric_div 10 --cider_weight 5 --data_percentage 100

Run run_all.py to obtain the evaluation from our proposed metrics and standard Recall. E.a. Evaluate the model on Flickr30K:

python run_all.py --model "$PATH_TO_SAVED_MODEL"model_best.pth.tar --split test --dataset f30k

E.a. Evaluate the mode with the full 5K testset of MS-COCO:

python run_all.py --model "$PATH_TO_SAVED_MODEL"model_best.pth.tar --split testall --dataset coco

If you found this research, metric proposed or code useful, please cite the following paper:

@article{biten2021image,
  title={Is An Image Worth Five Sentences? A New Look into Semantics for Image-Text Matching},
  author={Biten, Ali Furkan and Mafla, Andres and Gomez, Lluis and Karatzas, Dimosthenis},
  journal={arXiv preprint arXiv:2110.02623},
  year={2021}
}

Apache License 2.0