y-zheng18 / multihuman-dataset

Hi there 👋

About me: https://y-zheng18.github.io/

multihuman-dataset's Issues

Cropped Video Sequences

The video sequences provided for the Real World Capture dataset are cropped in the height dimension (i.e don't show the legs). On the other hand, the cropped images of each human show their full bodies. Is there a way to get the corresponding non-cropped full frames for all views?

Camera Calibration Files

In the Real World Capture dataset, the "calibration_new.json" files do not contain the distortion coefficients as part of the camera intrinsics. These coefficients are normally required by OpenCV to undistort the images as shown here https://docs.opencv.org/4.x/dc/dbb/tutorial_py_calibration.html. Should we assume that there is no distortion in the case of those videos?

SMPLX generation

Can you please share the method that was used to generate the SMPLX meshes for the given datasets?

How did you get the SMPLX of the MultiHuman datasets

I visualized MultiHuman/three/[500,502] sequence, it seems that some body parts of smplx are misaligned with the textured mesh. Did you use ICP or markerless mocap to get the SMPLX results?

dataset normalization

May I ask what operations are there for normalization, and what is the scaling ratio?

Mask Generation for Real-world Capture

I would like to ask whether the Segmentation masks (to segment humans from background) for the raw image sequences on Real World Capture are available or not? and if not any suggestions on how they could be reliably generated?

Weird Output When Generating 3D Keypoints

I was trying to compute the 3D keypoints and SMPL models for all the frames of the Real World Capture dataset, however, the result I got shows the humans having keypoints below the ground which lies at z=0. Does this mean that the camera calibration was done while assuming that the world center is not at [0,0,0]? If this is the case then where does the world center lie?

I've included the following visualization to show the mentioned effect.