https://github.com/darthzhu/vte

Source code of paper `Towards Visual Taxonomy Expansion`
https://github.com/darthzhu/vte

Last synced: about 1 month ago
JSON representation

Source code of paper `Towards Visual Taxonomy Expansion`

• Host: GitHub
• URL: https://github.com/darthzhu/vte
• Owner: DarthZhu
• Created: 2023-07-02T04:30:53.000Z (over 1 year ago)
• Default Branch: master
• Last Pushed: 2024-02-17T14:31:41.000Z (10 months ago)
• Last Synced: 2024-08-05T03:01:50.021Z (5 months ago)
• Language: Python
• Homepage:
• Size: 99.6 KB
• Stars: 6
• Watchers: 1
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• awesome-taxonomy - https://github.com/darthzhu/vte

README

        
# VTE

This is the official repository for *[Towards Visual Taxonomy Expansion](https://arxiv.org/abs/2309.06105)*.

## Dataset

### Chinese Taxonomy Dataset

The constructed Chinese taxonomy dataset is in `./Chinese taxonomy dataset`.

Due to commercial license, we cannot upload the images for each term.

However, it is possible to search each term on search engines to get access to its corresponding image.

In the training file, every hypernymy pair is stored as a list, where the first item denotes the hyponym, while the second item denotes the hypernym.

In the test file, the first item denotes the hyponym, while the second item denotes a potential parent derived from user click logs.

### Semeval-2016 Dataset

The original dataset can be view [here](https://alt.qcri.org/semeval2016/task13/index.php?id=data-and-tools).

The images used for this dataset are uploaded [here](https://drive.google.com/drive/folders/19dNsBkpxH4LD7ivz9toqNOMbv304edkI?usp=sharing).

## How to run

To reproduce our results reported in our paper, run:

```bash

python train.py \

    --train_datapath $TRAIN_DATAPATH \

    --dev_datapath $DEV_DATAPATH \

    --test_datapath $TEST_DATAPATH \

    --save_dir $YOUR_SAVE_DIR \

    --batch_size $BATCH_SIZE \

    --modal_integration add \

    --auto_add \

    --integration dot \

    --train_epochs $EPOCHS    

```