Published on 30 Jan. 2026
By Xiaoyuan Gao

I've always been interested in images. During my studies at ArtEZ, I did a project comparing how humans and machines describe unfamiliar images.  Maybe I'm biased—I find human descriptions more interesting, but what machines say is also quite special: very literal, and sometimes when they misidentify things, it's even more interesting. Last year I started messing around with image recognition and cropping again (mainly because I just really like collecting images). I downloaded a bunch of everyday photos from archive websites and cropped and categorized them very subjectively: birds, cups, how people look when standing, etc. Obviously this is some intense manual labor... I don't know what this attempt will become in the end, but I'll just keep doing it slowly. Below are some screenshots and categories. 

一直以来我都对图像很感兴趣，在ArtEZ上学时做过一个项目，对比人和机器怎么描述陌生图像。我比较偏心,觉得人的描述更有趣，但机器说的东西也很特别——非常字面,有时候识别错了反而更有意思。去年我又开始折腾图像识别和裁切（主要因为我就是很喜欢收集图像），从档案网站下了一堆日常照片，很主观地剪切分类：鸟啊、杯子啊、人站着的样子、等等。这显然是一个体力活……虽然不知道最后这个尝试会成为什么，我姑且就慢慢做下去好了。以下是一部分截图和分类。

Below are some cropouts from the YOLO model—it identified and extracted these from the archive photos.