easy12306

Why we included this project

easy12306 tackles the image-selection captchas on China's 12306 railway ticketing site, the ones that ask you to click every tile containing a given object. The author splits the problem into two neural models: one reads the caption text, like 'electronic scale,' and the other classifies each grid cell to decide which tiles match. A short command-line script then combines the two and prints the picks with a confidence score. The wiki and release history explain the design decisions along the way, including an earlier version that relied on similar-image search before the switch to convolutional networks. This is not production software, and you have to fetch the trained models from external links yourself, but as a compact, readable example of a real captcha being solved it is a handy reference for anyone curious about applied computer vision or how these puzzles get attacked.