search

Arthur151 / ROMP

Monocular, One-stage, Regression of Multiple 3D People and their 3D positions & trajectories in camera & global coordinates. ROMP[ICCV21], BEV[CVPR22], TRACE[CVPR2023]
https://www.yusun.work/
Apache License 2.0
1.34k stars 229 forks source link

3D model jittering #40

Open ahmedrasheed3995 opened 3 years ago

ahmedrasheed3995 commented 3 years ago

The model runs perfectly but I noticed there is a lot of jittering in the output. Is there any way to remove the jittering?

Arthur151 commented 3 years ago

Thanks for your attension. If you can obtain the tracking results of each person, you can take advantage of OneEuro Filter to smooth the mesh.

ahmedrasheed3995 commented 3 years ago

Thanks a lot for the reply, I'll try that. Meanwhile I tried applying savgol filter on estimated beta and pose values and it gave better results but only for single person. About tracking ,can you suggest which parameter I should track since there is no bounding box for persons. Thanks.

Arthur151 commented 3 years ago

For ROMP, each estimated body center location is pointed to each person. The body center is supposed to be the center of torso (L/R shoulders and L/R hips). You can associate the estimated body center with the tracking results from other tracking-by-detection methods.

ahmedrasheed3995 commented 3 years ago

Thanks a lot, will try that.

Gowan1998 commented 3 years ago

Hallo @Arthur151, thanks for the great project! I also have the problem with jittering. Can you tell me how to call OneEuroFilter class in test.py ? Do I need to pass outputs['verts'] to OneEuroFilter ?

Best regards

Arthur151 commented 3 years ago

Good question. If we want to optimize the temporal smoothness, we have to know the tracking results of multiple people. While currently released code lacks this part. We are developing this model. I will make sure it get released as soon as possible.

Gowan1998 commented 3 years ago

@Arthur151 perfect. I look forward to seeing more of this great project

Arthur151 commented 3 years ago

We have released the training code.