sanweiliti / egobody Goto Github PK

Hi,
thank you for your work, I find it very helpful to the community.

I would like to know if it is possible to know exactly which frames are part of EgoSet_interactee. Do you have a list? Or can I intersect some data somehow?

Thanks,
Alessio

Hi, thanks for your excellent work. Some of the files are too large (e.g., the size of kinect_color.zip is 352G) to download on my computer at the website. Could you please provide a download script or command line so that we can directly download the dataset on the remote server?

Hello, thanks a lot for your interesting work. Do you preprocess you2me dataset to compensate for fisheye distortions? Could you provide your evaluation scripts and fine-tuned network parameters?

I have registered and received the credentials. However, I can't see any download link information after logging in.

Thanks for your great work! It is a nice dataset with human poses in the world coordinate frame. In order to track the head-device wearer's motion, the IMU data can be very helpful. Could you please kindly consider releasing the IMU data as well?

Hi, thank you for your great work of EgoBody dataset!

I am doing some experiments which need other modalities of EgoBody dataset. Could you give an approximate time of when will these data be released? (Will them be released before ECCV 2022?)

I am looking forward to your reply!

Hello,

Thanks a lot for your great work. I wonder what the frame rate of motion capture sequences is, e.g. the sequence in:

smplx_camera_wearer_test\recording_20210911_S03_S08_01\body_idx_0\results

Thank!

Thanks a lot for your great work! I am wondering how u get the optical flow features. I think it is a little confusing. I want to evaluate the method on another dataset. I tried to extract optical flow through RAFT, followed by the pretrained Resnet18 to get of features from the optical flow images. However, the predicted head pose doesn't move obviously. So, I think I extract the features in a wrong way. May I know the details about how to get the of features?

Dear authors,

Great work! I'd like to ask the proper usage of 'scale' parameter which's defined in keypoints.npz.

Currently I'm using it the same way as mpii datasets. The corresponding bboxes are good in general, but lose track in some frames. For example, the thigh in this frame will be cropped out:

So I'm wondering if you have a proper guidance of 'scale' to crop the person out? Thx!