search

choyingw / SynergyNet

3DV 2021: Synergy between 3DMM and 3D Landmarks for Accurate 3D Facial Geometry
MIT License
383 stars 57 forks source link

Unsatisfied reconstructed results for images in-the-wild #17

Open zhanghm1995 opened 2 years ago

zhanghm1995 commented 2 years ago

Hi, thank your for sharing this amazing work.

When I ran the demo inference for the in-the-wild images, I got results they are not good.

The input image: image

The landmark image: image

The blended image: image

You can easily find the results have misalignment, I am not sure whether I did something wrong.

BTW, the image size is 512x512.

choyingw commented 2 years ago

image resolution and bbox cropping may affect the performance. You can try to tune the detected bbox size (increase or reduce the size a little bit) for the best performance.

zhanghm1995 commented 2 years ago

@choyingw Thank you for your reply. I want to know how to tune the detected bbox size with a little bit? BTW, the I found the image resolution of your sample images is also high, but the performance of yours seems better.

rlczddl commented 2 years ago

I encountered the same problem, how to adjust?

zhanghm1995 commented 2 years ago

@rlczddl Sadly to say, I cannot solve this problem.