7,000 Labeled Pokemon

### Context

I worked on a Pokemon classification project over the summer. (check out [my website](http://whatsthatpokemon.com) if you are interested, it's kind of buggy but mostly functioning) I couldn't find any usable datasets online so I made my own by going through every image in [this dataset](https://www.kaggle.com/thedagger/pokemon-generation-one), cropping each image, and removing falsely labeled ones , which was pretty painful and took me many hours but the results are great.

### Content

All the 150 Pokemons included in this dataset are from generation one. There are around 25 - 50 images for each Pokemon. All of them with the Pokemon in the center. Most (if not all) of the images have relatively high quality (correct labels, centered). The images don't have extremely high resolutions so it's perfect for some light deep learning.

在夏季，我参与了一个宝可梦分类项目。若对此项目感兴趣，可访问我的网站[whatsthatpokemon.com](http://whatsthatpokemon.com)，尽管网站存在一些小问题，但整体功能正常。由于网络上难以找到可用的数据集，我便自行构建了一个。我逐一检查了[该数据集](https://www.kaggle.com/thedagger/pokemon-generation-one)中的每一张图片，对图像进行了裁剪，并去除了错误标记的图片，这一过程颇为耗时且痛苦，但最终的结果令人满意。

数据集内容方面，包含了一代宝可梦中的全部150种。每种宝可梦大约有25至50张图像。所有图像均以宝可梦为中心。大多数（如果不是全部）图像质量相对较高（标签正确，居中）。图像分辨率并非极高，因此非常适合轻量级的深度学习研究。