jontooy/Flickr8k-Image-Features

---
language: ar
datasets: flickr8k
---

# Flickr8k Image Features

Flickr8k image features are extracted using the ResNeXt-152 C4 architecture ([found here](https://github.com/microsoft/scene_graph_benchmark)) and can be used as input for the [OSCAR](https://github.com/microsoft/Oscar) learning method. Arabic captions and splits are provided by [ElJundi et al.](https://github.com/ObeidaElJundi/Arabic-Image-Captioning)

## Dev-split
+ **dev-arabic.yaml** Yaml configure file with Arabic object tags
+ **dev.feature.tsv** Extracted image features
+ **dev.label.arabic.tsv** Arabic labels
+ **dev.label.tsv** English labels
+ **dev.yaml** Yaml configure file with English object tags
+ **dev_caption.json** Arabic captions for training
+ **dev_caption_coco_format.json** Arabic captions for validation

## Test-split
+ **test-arabic.yaml** Yaml configure file with Arabic object tags
+ **test.feature.tsv** Extracted image features
+ **test.label.arabic.tsv** Arabic labels
+ **test.label.tsv** English labels
+ **test.yaml** Yaml configure file with English object tags
+ **test_caption.json** Arabic captions for training
+ **test_caption_coco_format.json** Arabic captions for validation

## Train-split
+ **train-arabic.yaml** Yaml configure file with Arabic object tags
+ **train.feature.tsv** Extracted image features
+ **train.label.arabic.tsv** Arabic labels
+ **train.label.tsv** English labels
+ **train.yaml** Yaml configure file with English object tags
+ **train_caption.json** Arabic captions for training
+ **train_caption_coco_format.json** Arabic captions for validation

原始信息汇总

Flickr8k Image Features

数据集概述

Flickr8k图像特征是通过ResNeXt-152 C4架构提取的，可作为OSCAR学习方法的输入。数据集提供了阿拉伯语和英语的标注及分割。

数据集内容

Dev-split

• dev-arabic.yaml：阿拉伯语对象标签的Yaml配置文件
• dev.feature.tsv：提取的图像特征
• dev.label.arabic.tsv：阿拉伯语标签
• dev.label.tsv：英语标签
• dev.yaml：英语对象标签的Yaml配置文件
• dev_caption.json：用于训练的阿拉伯语标题
• dev_caption_coco_format.json：用于验证的阿拉伯语标题

Test-split

• test-arabic.yaml：阿拉伯语对象标签的Yaml配置文件
• test.feature.tsv：提取的图像特征
• test.label.arabic.tsv：阿拉伯语标签
• test.label.tsv：英语标签
• test.yaml：英语对象标签的Yaml配置文件
• test_caption.json：用于训练的阿拉伯语标题
• test_caption_coco_format.json：用于验证的阿拉伯语标题

Train-split

• train-arabic.yaml：阿拉伯语对象标签的Yaml配置文件
• train.feature.tsv：提取的图像特征
• train.label.arabic.tsv：阿拉伯语标签
• train.label.tsv：英语标签
• train.yaml：英语对象标签的Yaml配置文件
• train_caption.json：用于训练的阿拉伯语标题
• train_caption_coco_format.json：用于验证的阿拉伯语标题