dataset_info:
features:
- name: image
dtype: image
- name: filepath
dtype: string
- name: sentids
list: int32
- name: filename
dtype: string
- name: imgid
dtype: int32
- name: split
dtype: string
- name: sentences_tokens
list:
list: string
- name: sentences_raw
list: string
- name: sentences_sentid
list: int32
- name: cocoid
dtype: int32
- name: th_sentences_raw
sequence: string
splits:
- name: test
num_bytes: 819234726
num_examples: 5000
- name: validation
num_bytes: 807387321
num_examples: 5000
- name: train
num_bytes: 18882795327.165
num_examples: 113287
download_size: 20158273111
dataset_size: 20509417374.165
Dataset Construction
The dataset contructed from translating the captions of MS COCO 2014 dataset [2] to Thai by using NMT provided by VISTEC-depa Thailand Artificial Intelligence Research Institute [3]. The translated of 3 splits (train, validation and test) dataset was published in the Huggingface.
References
[1] C. Polpanumas and W. Phatthiyaphaibun, thai2fit: Thai language Implementation of ULMFit. Zenodo, 2021. doi: 10.5281/zenodo.4429691.
[2] Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C. Lawrence Zitnick. 2014. Microsoft COCO: Common Objects in Context. In Computer Vision – ECCV 2014, Springer International Publishing, Cham, 740–755.
[3] English-Thai Machine Translation Models. (2020, June 23). VISTEC-depa Thailand Artificial Intelligence Research Institute. https://airesearch.in.th/releases/machine-translation-models/