Hi all! I’m having an issue when using mesh to metahuman with a scanned mesh from reality capture. It looks like the eye trackers are way off the mesh causing the eyes to bug out all crazy!
Data driven eyes is ON… when I turn it off the eyes are too small and sunken into the head.
Is there a work flow to fix this? It seems like you need to make adjustments to the trackers in the front view with no way to adjust their depth/offset from the surface?
Thanks @mcleary! Totally, that is what I did in the last instance (promote the front view, track and hit identity solve) I just rotated to another angle after the fact to show the trackers were off the mesh.
I just tried it again with a fresh metahuman identity node, I followed all the recommended steps… cam fov 20, unlit, fill most of the frame straight on, activate trackers, identity solve with data driven eyes OFF… and this is the result. The eyes are small and sunken into the head. Is there any way to adjust the eye placement before hitting mesh to metahuman? How can I fix the eyes… I searched on here and people suggested turning on data driven eyes however that results in crazy bug eyes
I work on the Core Tech side of MetaHuman Animator. It looks like, for some reason, the facial tracking points are sitting in the wrong place in 3D. Just to confirm, as Thales said, if you promote a single frontal frame, are the tracked facial points sitting in the wrong place in 3D? That will be what is causing the ‘bug-eyes’. It is clearly a bug!
Could you send a screen-shot of front and side views of the frontal promoted frame?
Hi @JaneHaslam ! thanks for looking into this. What is the correct workflow to see the trackers from the side after you promoted the frontal frame and activated the trackers?
It seems if you unlock the camera and rotate the frame the trackers disappear.
Hi @mcleary ! Ok here is the result… at first the iris trackers look ok but the eye placement on the template mesh is off… then when I unlock and rotate the view the iris trackers are in the off position
Hmm. Looking at the video, it does look like you are using the tool correctly and the annotations are correct; it also looks like there is good mesh data in the eye region. Mesh 2 MetaHuman does rely on having a good quality mesh. I am wondering if the fact that the facial input scan does not cover the full frontal part of the face is causing this. Is there any way to create a slightly more complete facial scan?
I also have one final question. Is the facial scan ‘metric’ scan ie the scan of the scan is correct? Mesh 2 MetaHuman does rely on the input data being within a reasonable range of scale or size for an actual human face/head.
I think to investigate further, we would have to debug and investigate using the actual scan data.
The input data looks reasonably sensible. The scale which is coming back from the solve is quite large (1.5 x mean head), but I don’t think this is the issue, as rescaling during input is not helping. Reality Capture scans are not guaranteed to be metric scale I believe.
I can’t find an immediate work-around I’m afraid but will report it as a bug which needs investigation.
@JaneHaslam I’m glad you’re looking into this. I’m running the latest UE 5.2, and the latest Metahuman plugin updates, and I have a really nice-looking photogrammetry head that is doing the exact same cross-eyed look. I’ve made many of these heads using this technique with no issues until now. I hope you can figure out the issue!
I was wondering if this was a 5.2 issue. I’ve tried 4 different heads and they all have wonky eyes. I’ve never had this problem until now. It’s frustrating.
I have never seen this issue on both the previous versions 5.0 & 5.1. Mesh is complete, scaling is proper and ever since 5.2 this issue persists and doesnt go away, I tried with many characters. Sometimes data driven eyes gives a temporary solution but doesnt work for all. PLEASE LOOK INTO IT AND FIX IT.
As you can see, I am having the exact same issue. I am using 5.2. There were some minor errors in 5.1 regarding the eyes in particuar, but nothing of this magnitude. Does anyone know what is causing this?