Some wrong with the results of Novel View Synthesis

zju3dv / EasyMocap

Make human motion capture easier.

Other

3.43k stars 439 forks source link

Some wrong with the results of Novel View Synthesis #284

Open Haza628 opened 1 year ago

Haza628 commented 1 year ago

Hi! I used a three-person scenario dataset for training, and the training results are shown in the figure. depth_map_000029 feat_map_000029 instance_map_000029 rgb_map_000029

Why does everyone's render result have a bounding box？ Appreciate any help!

chingswy commented 1 year ago

Seems that the person stands on their own place and doesn't move. The network fails to decompose the background and the foreground.