README.md

metadata

license: apache-2.0

[EMNLP'23] ReSee: Responding through Seeing Fine-grained Visual Knowledge in Open-domain Dialogue

ArXiv: https://arxiv.org/abs/2305.13602

Code: https://github.com/ImKeTT/ReSee

This is the processed data for ReSee, more raw image data are coming...

The data should looks like this:

.
├── ./processed_resee_data  
    ├── dd # Contains proccessed entity-level image features and annotations of DailyDialogue
        ├── processed_img_features
            └── img_clip_features.pt
        ├── test_v0.json
        ├── valid_v0.json
        └── train_v0.json
    ├── wow # Contains proccessed entity-level image features and annotations of Wizard of Wikipedia
        ├── processed_img_features
            └── img_clip_features.pt
        ├── test_random_v0.json
        ├── test_topic_v0.json
        ├── train_v0.json
        ├── valid_random_v0.json
        └── valid_topic_v0.json
    └── shared # Turn-level image features
        ├── coco
        ├── flickr30
        ├── nocaps
        ├── openimagev6
        ├── processed_img_features_clip_base # turn-level image features processed by ViT base
            ├── coco_train_clip_vis_fea.pt
            ├── coco_val_clip_vis_fea.pt
            ├── flickr30_clip_vis_fea.pt
            ├── nocaps_clip_vis_fea.pt
            ├── openimagev6_test_clip_vis_fea.pt
            ├── openimagev6_train_clip_vis_fea.pt
            ├── openimagev6_val_clip_vis_fea.pt
            └── oodcv-counterfactual.json
        └── processed_img_features_clip_large # turn-level image features processed by ViT large
            ├── coco_train_clip_vis_fea.pt
            ├── coco_val_clip_vis_fea.pt
            ├── flickr30_clip_vis_fea.pt
            ├── nocaps_clip_vis_fea.pt
            ├── openimagev6_test_clip_vis_fea.pt
            ├── openimagev6_train_clip_vis_fea.pt
            ├── openimagev6_val_clip_vis_fea.pt
            └── oodcv-counterfactual.json