Closed xungeer29 closed 8 months ago
Our HaMeR network estimates the hand pose from the local bounding box of the hand. The translation is inferred as a function of the provided focal length. The demo uses a large focal length that approximates weak perspective projection. If you use a more reasonable focal length value (e.g., f=1000), you will get something that looks like this:
Please check also the explanation in this issue.
Thanks for you great work! The rendering result is very good, but why is the relative position between the hands so poor?
![image](https://github.com/geopavlakos/hamer/assets/32890569/0ae62efc-ce0a-4ef9-bea6-f650d18e0026)