Sunday, March 4, 2018

Brandon Wehenkel: Independent Study (Update 2)




Comparison

Since I already established the purpose of this study here, I won't need to reiterate the exact same thing again for this and the next two posts.  Unfortunately, by the time I created my first blog post I was already 3 weeks behind.  The scanning process was an utter nightmare for me, and I've understood that I may have made my sculpture a bit too detailed (I still wanted more detail too) and too large for the scope of the NextEngine 3D scanning device.  This time around I spent many hours after normal business hours in the lab working in the RapidWorks application.  The first week of February was nothing but hole filling and retopologization for the standing figure only.

For a turntable go here.

Modeling Techniques:

RapidWorks, Body
A great tool I used a lot was sew boundary, it fixed a lot of the Zfighting.  Healing Wizard also helped me identify which parts of my model had issues, not everything was fixed by the wizard, most of the time it was an assistance tool.  Isolating the problems that were identified with the healing tool made it easier to clean the mesh.
Standing Figure 
Before Alignment, 16M polys
Here you can see the color variation for each layer that became a clutter of massive problems.  A whole lot of messed up geometry and misalignment issues.  I also had an issue with the texture not following through to RapidWorks as well, but that was the least of my issues.

After Alignment, 26.85M polys
By the way, aligning took 6 hours with Rewrap.  The solution to reducing time, smooth = fewer polys.  I used three different methods:  Mesh Buildup Wizard, Rewrap, Optimize Mesh.  Optimize mesh never worked, it was an absolute 100% waste of my time.  From 1pm to 10pm the computer did nothing but "load," at one point the computer was running at 92% memory capacity due to this tool.  In my case, the wizard helped tremendously while the rewrap helped fix some of the overlapping geometry.  However, I still had to go in and delete a lot of the manifold edges and overlapping geometry because they made the model mucky like chunky peanut butter.  Some areas such as the buttocks and the legs were so destroyed by the manifold faces that I needed to rebuild a lot of the modeling with the sandpaper tool by hand.


One of many massive issues that came with Rewrap (after Global Remesh retopologizing) and decimate is that a large chunk of the normals flipped.  I am using the bridge tool here.  Take note of the massively broken boundary (aka squiggly yellow line on the back of the head).  Most of the model looked like this.  The horns were destroyed entirely from underneath.  The face lost a lot of the modeling from the nostrils to the neckline under the chin.  The gap underneath the udder to the pelvis was nonexistent and contained a boundary that went from the hand all the way down to the knee.  All the effort I put into making a unique stand also went to waste, now it's mostly flat.  A large chunk of that boundary was created from one of my partial scans from ScanStuido. (Later, I had to rebuild the stand in Mudbox just because RapidWorks didn't create a nice smooth mesh).

IMPORTANT INFORMATION: If you're trying to fill holes in ScanStudio and have an issue with the model being too large for the view window and think that you'll just fill the gaps with partial scan data, don't do it. It's best to not do close up scans of the object that are only partially aligned.  Make the decision to fit the entire object within the window or pick another scanning software.  If you can't make a fully attached scan discard it, otherwise you'll end up with a lot of massive broken boundaries in Rapidworks when trying to fill holes. Or unattached scan data.  The reason I chose to make only partial scans is that I couldn't get the torso to fit in the view window, so I let the scanner do what it needed to do and moved on with the data that I had.  Because I had about 1/3 of the torso scanned, the other 2/3rds were broken.  The location where the extra scan data overlapped with the original mesh caused a lot of flipped normals and massive gaps in geometry.  The tools in Rapidworks that try to remap or retopologize the sliced date sew pieces that you might not want to be sewn together creating a lot of problems later.


PARTIALLY ATTACHED SCAN DATA IS BAD!!
Scans B, C, and D were only sections of data, which basically corrupted my geometry.

 
This is the best example of what I'm talking about. (Ignore the massive wall on the backside of the figure, that was something that came about from using the wrong setting in a tool.)  The left thigh has a massive flipped normal, that is not a hole.  There are two layers of data on this model.  The legs had a more clean set of geometry, and the geometry underneath caused a lot of that messy, jagged noise.  This was a nightmare to repair.  Also if you go back to the image that I posted above with 26M polys, you can see the floating geometry that caused a seam from from the pelvic region all the way to the tip of the ear.  Most of it was deleted.  The hands were never fixed entirely in Rapidworks.

Mudbox, Body
I imported the body .obj file into Mudbox. Before you go any futher make sure you hold "spacebar" and select the arrows to change the position and scale of the object.  The importing process into Mudbox is dumb and the models are always scaled wrong.  When I imported the models, there were a few errors such as High-valence vertex, Two-sided faces, Zero-length edges, Degenerated Face. The best thing to do to correct these:
  1. Make sure you're in triangular topology (quads don't work, *go here for more info)
  2. Press "V" (goes into selection mode.)
  3. Select all of the faces surrounding the error.  
  4. Delete. 
  5. Then use Mesh --> Patch.
  6. Undo and redo the process until this works.

 Holes

I had run into a stopping point for the week of February 19-25th because of a significant error that has prevented me from continuing.  Sometimes on 22nd, I noticed I had a hole in my mesh post-quad topology.  Some reason Mudbox doesn't like editing errors in this topology type, so I imported the file into Rhino to clean the holes and discovered there's a vertex alteration in the .obj conversion.  The data becomes 4x larger when importing from Rhino to Mudbox.  The file never loads in Mudbox, unfortunately.  Professor Scott said to ignore the holes till I need to work in Rhino officially.   The Feb 23- Mar 3rd I was busy cleaning up the butter mesh on the body.

* March 5th, I discovered the solution to this hole problem and it was much easier to fix than I had thought previously. Use the Freeze tool to mask the holes (to prevent tearing in the mesh) then use the Remesh tool under Mesh --> Remesh, similar to Smooth --> Decimate in Rapidworks.  It asks for a percentage and poly count.  Just type in the poly count and it retriangulates the model.  Then you can go back and select all the polys around the hole with the "F" key and select mesh --> patch.


 Goat progression


The feet progression.

Because the feet were made on the base, the scanner didn't really like it much.  So I had to rebuild the feet twice.  Once in Rapidworks and again in Mudbox.  Nothing different was involved with Mudbox, I used all the same tools as the rest of the body.
 

Sitting Figure:

Import from ScanStuido

Udder mess

Scan Studio, Sitting
This part of the figure is the sitting portion of the process.  I did not run into any significant issues with this part.  The figure fit perfectly centered in the view port, and I only needed to do 2 scans, a top, and a side scan.  I aligned them with minor errors.

Before cleaning

Bridging the hoof
RapidWorks, Sitting
I first I fixed the viewport alignment with tools --> align --> Datum and changed the right image to match my view window for the front.
Then I used Mesh Buildup Wizard to attach all my scans from scanStuido to make a workable mesh.  The main problem is that the dew claws and the forward section of the hoof were misaligned similar to the standing figure, so I spent most of my time in RapidWorks cleaning up those areas.


Cleaned version 1

I had a few crazy boundaries that required some finesse.  Like I mentioned before the areas around the hooves were the worst.  I had to destroy most of the exterior mesh to rebuild it with the fill holes tool, golf and bridge were most useful.  I don't much like plateau much because it doesn't really know which part of the mesh to destroy; sometimes it erases the entire mesh.

The groin area was not possible to fix in RapidWorks.  I had to adjust in Mudbox.  There was a lot of non-manifold edges there, and I did the same as the hooves.  Used bridge and golf to remake the modeling of the legs and the teets.  The legs are folded over one another in this location and because there isn't a "pinch" tool like Mudbox/Zbrush.  In this program, you have to trick the mesh by building and destroying the mesh over and over till you get the right elevation for the geometry.


Mudbox, Sitting
The udders, the modeling around the buttocks and ankles, the creases, and the hooves were lost in the scan.  Same as the body and the arm I used the smooth, bulge, and knife tools. 

Cleaned version 2


Arm:
Mudbox Cleaned


Because all the issues I had with the standing figure I needed to create a new arm.  The mesh for the arm was fused to the udder, and I need to create a part that could be separated from the rest of the mesh for when the goat is sitting.  The standing figure is holding their udder while the sitting figure is not.

Scan Studio, Arm
Only one full 360 revolution scan was required for this model. I used 6 photos to complete the scan due to its simplicity.  It's just an arm.

RapidWorks, Arm
Same as the last two meshes, I first uploaded the scanStudio file then used mesh buildup wizard to attach all my scan data.  The bottom had a hole, and it was patched with bridge and golf.

Mudbox, Arm
Imported to Mudbox where the smooth, bulge, flatten, and knife tools were used significantly.

Rhino, All
I imported all the models into one file and manipulated them to begin my cutting.  I put all object on the origin by using Move and typed 0,0 in the toolbar.  Then used the various perspectives to position each object to intersect its corresponding geometry that I want to cut.


At this point, I wanted to see how the geometry matched up so I can go back and forth on Mudbox & Rhino.  Before and after, I wanted to bulk up the appendages. I made a minor error when creating the extra arm, so thats the last thing I need to fix before I cut all the geometry for 3d printing.

Problems:
This project was much more ambitious than I thought.  I am seeing that the descrepancies between the scan and the next stage in Mudbox/Maya/Rhino is a significant amount of repair and remodeling time.  Because Rapidworks doesn't have much for modeling a lot of gemoetry is lost when repairing the alignment problems.  I spent several hours just working in Mudbox which I didn't expect was going to happen.

The hardest part for this project is correcting my own bad geometry.

No comments:

Post a Comment