#PraCegoVer dataset

Automatically describing images using natural sentences is an essential task for visually impaired people's inclusion on the Internet. Although there are many datasets in the literature, most of them contain only English captions, whereas datasets with captions described in other languages are scarce. PraCegoVer arose on the Internet, stimulating users from social media to publish images, tag #PraCegoVer, and add a short description of their content. Inspired by this movement, we have proposed the #PraCegoVer, a multi-modal dataset with Portuguese captions based on posts from Instagram. It is the first large dataset for image captioning in Portuguese with freely annotated images. #PraCegoVer has 533,523 pairs with images and captions described in Portuguese collected from more than 14 thousand different profiles. Also, the average caption length in #PraCegoVer is 39.3 words and the standard deviation is 29.7. <strong>New Release</strong> We release <strong><em>pracegover_400k.json</em></strong> which contains 403,337 examples from the original <strong><em>dataset.json</em></strong> after preprocessing and duplication removal. It is split into train, validation, and test with 242036, 80628, and 80673 examples, respectively. <strong>Dataset Structure</strong> #PraCegoVer dataset comprehends a main file <em>dataset.json</em> and a collection of compressed files named <em>images.tar.gz.partX</em><br> containing the images. The file <em>dataset.json </em>comprehends a list of JSON objects with the attributes: <em>user: </em>anonymized user that made the post; <em>filename: </em>image file name; <em>raw_caption: </em>raw caption; <em>caption: </em>clean caption; <em>date: </em>post date. Each instance in dataset.json is associated with exactly one image in the <em>images</em> directory whose filename is pointed by the attribute <em>filename</em>. Also, we provide a sample with five instances, so the users can download the sample to get an overview of the dataset before downloading it completely. <strong>Download Instructions</strong> If you just want to have an overview of the dataset structure, you can download <em>sample.tar.gz.</em> But, if you want to use the dataset, or any of its subsets (63k, 173k, and 400k), you must download all the files and run the following commands to uncompress and join the files: <pre><code class="language-bash">cat images.tar.gz.part* &gt; images.tar.gz tar -xzvf images.tar.gz</code></pre> Alternatively, you can download the entire dataset from the terminal using the python script download_dataset.py available in the PraCegoVer repository. In this case, first, you have to download the script and create an access token here. Then, you can run the following command to download and uncompress the image files: <pre><code class="language-bash">python download_dataset.py --access_token=&lt;your access token&gt;</code></pre>