Ongoing Xiahou Ba Rigging Failures

Bare in mind before reading this, this is the 3rd time I've pretty much typed this issue out trying to get help to fix it from Polycount and Area forums.

The issues are less than when I previously posted thankfully, however they are still far from correct. 

The stage I am at now is simply putting the motion capture skeleton into T-pose to match my characters pose, keyframed on frame -1 so that it doesn't affect the data in any way. This T-posed version is the duplicated to make sure both rigs are exactly the same, but of course the duplication does not keep its keyframed animation data.

Using Maya's built-in HumanIK system for retargeting animation data, I create a new skeleton rig for each skeleton. Then you simply tell the character's skeleton (the duplicate), to follow the original motion capture skeleton as a source of data. This works, but goes very wrong.

Above shows both the original and duplicate rig on top of one another. I have also told the major joints to show their local rotation axis to show that they are set up correctly and are both identical to one-another. You can also see HumanIK believes there to be no problems with either skeleton.

Above here shows you what happens when I tell my character's skeleton to follow the motion capture data. The skeleton of my character becomes all bent out of shape. Below shows you a better rendition of what is going wrong here:

This is very frustrating. The issues looks like it must be caused by a difference in joint rotation between the two skeletons, but there isn't one. I'm not sure why the skeleton moves its root (pelvis bone) up a bit from where it should be too, or why the wrists are bending when the animation is played instead of the elbows as they are on the motion capture data. Both skeletons are set up anatomically correctly in the HumanIK editor, so surely this should not happen either.

All I can do is wait for someone with better knowledge in this to get back to me and hope its a simple fix. So close, yet so far.... I must be missing something really silly!

Xiahou Ba Progression - Difference Between Baking Normal Maps in ZBrush & Maya

While I continue to develop my rendition of Xiahou Ba from Dynasty Warriors 7/8, I have been running some tests with the texture data I'm getting back from the high resolution sculpt in ZBrush. The UVs are holding up very well which I am pleased with, but I noticed some of his normal maps have a little to be desired still; they often seemed a bit faint or flat in areas, compared with the true 3D sculpture.

Looking into this problem online for a while, I started reading posts from people saying ZBrushs' normal map baking is very weak and they don't even use it for this reason. Instead, they use their original 3rd-party application for this job, such as Maya, Modo, Max, or XSI.

(Softimage XSI no longer exists as of 2015 by the way)

With this idea, I decided to bring a high resolution mesh from ZBrush (Sub-division Level 5) into Maya, along with its 'game-ready' counterpart. I then used Maya's 'Transfer Maps' tools under its 'Rendering' tool-set to bake the surface information from the high resolution mesh to the low resolution mesh in the form of a normal map. The results certainly are different, and it does seem Maya does produce slightly superior results:

This GIF animation shows the actual difference between the bake from ZBrush, compared with the bake from Maya. The frames are spaced 2 seconds apart, so give it time.

To accompany this, I will provide you with the high resolution bakes themselves. The top image shows the results split next to eachother, and the bottom images show the full results side-by-side.

From studying the differences between the two map bakes, both completely unadulterated, you can make some general observations straight off the bat:

• Maya's bake contains stronger reds and greens, increasing the overall intensity of the normal map compared to that from ZBrush.
• ZBrush by enlarge has a cleaner resulting map straight away, with less tangent distortion around the neck and bottom of the torso on the map. However it is important to point out at this stage that Maya can fix these issues with tweaking of its options; it is simply the case that ZBrush appears to be more optimised for this kind of work from the get-go.
• ZBrush appears to have possibly distorted the baked results very slightly, most noticeable on these bakes around the nipples and under the arms. Despite saying that however, I'm still not 100% sure it really is distortion, and it could be Maya that has distorted the results instead and that they simply just look a bit better. This would all be down to the target search area set up in each software package accordingly.

In other words, it might be that Maya produces better, stronger results overall, but they require more tweaking around the edges and more tweaking of the options before each bake than ZBrush. It would theoretically be possible to also intensify and adjust the normal map bakes from ZBrush to get better results also, however this process can be very difficult to do without breaking the continuity of the maps (seams can suddenly become violently visible if you simply increase the contrast on the maps).

These GIF animations shows the results working on the actual model. Especially pronounced on the greyscale images, you can see that the baked map from Maya gives a much greater sense of 3D than that from ZBrush.

I would not call this testing conclusive. While Maya holds better results than ZBrush for this model, the same might not be true for other models. Also remember that both of these maps are made using somewhat standard settings in both applications, and so further tweaking of the options may produce better results and worse results in both software packages respectively. However the issue with ZBrush baking normal maps to look much flatter and less intensive is something that has no options to modify. For this reason alone, I think further testing with Maya might hold better results for my future projects.