About dataset annotation

[Followmeczx]

I looked at the annotated json files for the MSCOCO, MPII, Human36M datasets. I noticed that none of the camera parameters had extrinsic parameters for rotation and translation, but only intrinsic parameters for focal length and principal point. Why is that? What if I want to get the camera calibration parameters? Can you give me some advice

[linjing7]

Hi, we use the weak perspective camera model. You can refer to this paper for more detail.

[pasin-k]

Hi, I wonder if I have my own image which I would like to add to train, how can I find out intrinsic parameters?

@linjing7 In the paper you mentioned, it mention assumption of initial focal length = 5000 but does not mention how it scales with cropped image.

I checked UBody dataset and still find no correlation on how each object has its own focal length value. Any documentation of how the focal length is calculated will be much appreciated.