apple / ml-hypersim

Hypersim: A Photorealistic Synthetic Dataset for Holistic Indoor Scene Understanding
Other
1.71k stars 131 forks source link

Opencv format Camera Intrinsic #72

Closed lqql2012 closed 4 months ago

lqql2012 commented 4 months ago

Thanks for your great job , As answered in question #44 , the intrinsic parameter fx of the dataset is 886.8,The format written in opencv may be [[ 886.8,0,512],[0, 886.8, 384],[0, 0, 1]] ,How is 886.81 calculated?

mikeroberts3000 commented 4 months ago

I don't understand the question. Can you re-phrase it in a more technically precise way? Where does 886.81 come from? Where do you see that number?

lqql2012 commented 4 months ago

I don't understand the question. Can you re-phrase it in a more technically precise way? Where does 886.81 come from? Where do you see that number?

I saw this data here: https://github.com/apple/ml-hypersim/issues/44#issuecomment-1082702513 .I need to train a model across multiple datasets ,the camera intrinsic parameters of other datasets are in pixels , such as NYUv2 、DIODE and so on. Their focal length is calculated as : focal length [pixels] = focal length [mm] / sensor pixel size [µm/pixels]

lqql2012 commented 4 months ago

My question same as : https://stackoverflow.com/questions/78072261/how-to-find-cameras-intrinsic-matrix-from-focal-length

mikeroberts3000 commented 4 months ago

You still haven't stated your question in enough technical detail for anyone to help you.

The only concrete question you have asked so far is in your first post, where you asked about a specific scalar value that doesn't appear anywhere in our code. Your second post doesn't ask a question, and doesn't provide any meaningful clarifications. And your third post simply links to an already-answered Stack Overflow question with no other context.

This isn't helpful. You will get more effective help on GitHub and in other venues if you learn to ask questions more clearly. For example, if your question is actually the same as the one you linked to on Stack Overflow, then why doesn't the accepted answer on Stack Overflow work for you? My main suggestion would be to slow down, and write out your question in a clear methodical way. Your goal is to make your question make sense to a reader that isn't familiar with exactly the same mathematical notation and mathematical conventions as you.

With all of that being said, we provide the field-of-view for every Hypersim scene, so you can use basic trigonometry to calculate focal length in whatever units you want. Think of the image plane as the base of a pyramid, think of the field-of-view as the angle at the top of the pyramid, and think of the focal length as the height of the pyramid. Once you have chosen the size of the pyramid's base, there is only one possible height for the pyramid that matches the angle at the top.