Get the latest tech news
Line scan camera image processing for train photography
I use my line scan camera to take cool pictures of trains and other stuff.
Generally, I do a decent job of keeping the camera upright, so we just need to correct for very small skews, so a Hough transform is suitable (since the complexity scales with the number of bins of the histogram). For example, when trying to implement spline-based resampling, ChatGPT 5 came up with horribly slow (but vectorized) code that constructed a giant tensor with a mask across the entire width of the image for every single sample. Both Grok 4 Expert and ChatGPT 5 Thinking also completely failed to implement the band-diagonal least squares to my vertical stripes strategy, but as mentioned, the exponential smoothing trick works okay for now.
Or read this on Hacker News
