search

microsoft / multiview-human-pose-estimation-pytorch

This is an official Pytorch implementation of "Cross View Fusion for 3D Human Pose Estimation, ICCV 2019".
MIT License
541 stars 89 forks source link

Questions about the human3.6m annotation file #2

Closed Ixiaohuihuihui closed 5 years ago

Ixiaohuihuihui commented 5 years ago

Hi, First of all, thanks for open sourcing this great work. I have some questions: (1) This work retrain 2D pose estimation network using 2D ground truth heatmap of Human3.6 dataset, right? Because other works only use state-of-the-art 2D human pose estimator to generate 2D labels for Human3.6M without using the 2D ground truth. (2) Another question is that how to generate the h36m_train.pkl？The file contains "joints_2d" and "joints_3d", how to combine the image file with corresponding joints 2d or 3d information? Thanks in advance!

CHUNYUWANG commented 5 years ago

(1) Yes, we train the whole network, including the fusion parameters, on the H36M dataset.

(2) The 2D and 3D joint locations are provided by the dataset. We will release a toolbox on processing the H36M dataset soon.

Ixiaohuihuihui commented 5 years ago

Thank you! Looking forward to the releasing news!

CHUNYUWANG commented 5 years ago

Please check this repo for preparing images and annotations https://github.com/CHUNYUWANG/H36M-Toolbox

Ixiaohuihuihui commented 5 years ago

Thank you very much!

lisa676 commented 4 years ago

@CHUNYUWANG Hi, I've npz and h5 format for Human3.6m dataset. IS there anyway to convert these format into pkl format?

Ixiaohuihuihui commented 4 years ago

@ahsan3803 Hi, I think the convert progress is simple. Just read the array from h5 file and write the array into pkl. For example:

import h5py
with h5py.File('../3d-pose-baseline/data/h36m/S1/MyPoses/3D_positions/Directions 1.h5', 'r') as h5f:
    poses = h5f['3D_positions'][:]
with open('h36m_train.pkl', 'wb') as f:
        pkl.dump(poses, f)

However, the key issue is that how you get the same data keys in your annotation files...

lisa676 commented 4 years ago

@Ixiaohuihuihui Hi, thanks so much for your kind response. Yeah you are right it's key issue. We can't input NPZ file also due to 2D and 3D joints/pose information. Am unable to get Human3.6m original dataset, I tried many times but my account is not verified yet. Is there any other way to get this dataset?

Ixiaohuihuihui commented 4 years ago

@ahsan3803 Send an email to Human3.6m team. And you need to supply any information which can prove your relationship with your advisor.

zhhaochen commented 4 years ago

@ahsan3803 Send an email to Human3.6m team. And you need to supply any information which can prove your relationship with your advisor.

Hi, I send an email to Human3.6m team but no reply. Can you share the pkl files? Thank you!