HybrIK

HybrIK copied to clipboard

hi, how to export hybrik to onnx?

lucasjinreal

Computation of ground truth twist angle

1

Hey, the issue "How to make annotation files" already relates to my question but my questions rather specifically directs to the computation of the twist angle: Did you compute the...

odunkel

您好，有个问题想请教下 我这边在重新训练模型时，预加载pose_hrnet_w48_256x192.pth权重，同时加入了新的数据集，导致的结果是在新数据集上表现变好，但是pw3d和h36m数据集的MPJPE分别增加了20到30mm 于是我做了另一个实验，只用原有的数据集训练，配置文件如下，最后得到的指标依旧达不到pose_hrnet_w48_256x192.pth那样好的效果，在零散的测试视频上也没有pose_hrnet_w48_256x192.pth效果好，想请教下是否在训练数据集设置或者损失函数权重上有需要注意的地方，或者需要调整的地方，如能有空解答，非常感谢！ DATASET: DATASET: 'mix2_smpl_cam' SET_LIST: - ROOT: './data/h36m/' TEST_SET: 'Sample_20_test_Human36M_smpl' TRAIN_SET: 'Sample_trainmin_train_Human36M_smpl_leaf_twist' - ROOT: './data/coco/' TRAIN_SET: 'train2017' - ROOT: './data/3dhp/' TRAIN_SET: 'train_v2' - ROOT: './data/pw3d/' TRAIN_SET: '3DPW_train_new' PROTOCOL:...

tucker666

I ran the demo_video.py but I got the following error ``` Traceback (most recent call last): File "scripts/demo_video.py", line 137, in hybrik_model = builder.build_sppe(cfg.MODEL) File "/home/yqzhang/HybrIK/hybrik/models/builder.py", line 21, in build_sppe...

qiqiApink

Extracting 3D model

2

Hi Jeff & HybrIK User Group, Can you kindly suggest the best way to export the animated 3D mesh after HybrIK has completed? Currently, I'm only able to export a...

asicdesign

demo_video.py error

2

Loading model from D:\PythonProjects\HybrIK\pretrained_w_cam_res50.pth... Traceback (most recent call last): File "D:\PythonProjects\HybrIK\scripts\demo_video.py", line 101, in hybrik_model.load_state_dict(torch.load(CKPT, map_location='cpu'), strict=False) File "C:\Users\hanzhao\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 1667, in load_state_dict raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(...

zhohn0

 I use pip install -e . , but it is invalid. Finally , I try using # Anaconda Cloud conda install pytorch3d -c pytorch3d , it works!

Ethan-cpp

Hello, me again ^^. Here I wondoer why do you estimate camera parameters instead of the one provided in dataset directly?https://github.com/Jeff-sjtu/HybrIK/blob/aa1ef5e8dbe004d70e545569b261e4a2bdfe8988/hybrik/utils/presets/simple_transform_3d_smpl_cam.py#L352

PGogo

Congratulations to the authors for this great research work! I am trying to understand the generation process of 'thetas' and 'betas' parameters of Human3.6m. I cannot download the Mosh code...

moezet01

无法重现b站效果

3

你好，我在b站看到up整活，效果很好，但我实际测试却是无法是识别到人 

LaolangOne