CAN NOT be applied to real-world situations

[TekamanBlade]

For this method to achieve accurate hand pose estimation under occlusion conditions, it first requires an accurate bounding box to cover the complete hand area. However, the problem of obtaining the hand area under occlusion conditions itself demands the ability to accurately estimate the hand pose under occlusion. Therefore, this method simply cannot be applied to real-world scenarios. Am I right?

[dqj5182]

As you noted, 3D hand reconstruction has two main components of 1) Hand detection, 2) 3D hand mesh reconstruction. These two components are separately studied so that each study can focus on task-specific challenges.

It is true that this method cannot be directly applied to real-world scenario.

But, in my opinion, this method gave good perspective on how to contemplate on occlusion.

If you need code for real-world scenarios, I recommend using recent methods like WiLoR that jointly conducts hand detection and reconstruction.