CLEAN UP - 3D PROJECTIONS
Clean up for music video "Grand" - directed by Jake Yates, Post Production by Square Pixel Films
From our FRAMESTORE Feedback session with ARIEL LEVENTAL I learned something I did instinctively:
REPLACE JUST WHAT YOU HAVE TO REPLACE!
01 SHOT SELECTION
After selecting the first shot, I noticed I had to do a lot of roto work, so I decided to change shot. In the beginning, I was planning to remove people from the shot I choose. Than Square Pixel had a shot to work on for a music video. I wanted to take the opportunity to work on a real project!
02 ANALYSE
Analyzing the shot! Watch and rewatch the shot, analyze what exactly has to be done. Once you start, don't forget to replace just what you have to replace. Try to find out the camera information and lenses, which were used for the shot.
03 TRACKING
Tracking for the clean up shot is ESSENTIAL. After changeing and working on different shots, I learned how IMPORTANT it is to have a GOOD TRACK. Once you have a good track, you have done half of your clean up.
04 LENS DISTORTION
Lens distortion is quite important to get a decent track. If possible, you should know which lens was used for the shot as it makes it easier to track later. This link can help to find camera and lens information: VFX Camera Database
05 GRAIN
Using the F_Grain node gave me the best result: When regraining, ideally place the F_Grain over the same area where the denoise was placed.
OH BOY!!!
Problems:
01 TRACK
Knowing that the track is one of the most important thing for the clean up, I had to track and retrack the footage a few times. After tracking I placed some cards, to see if they stick. After the first track, my card was sliding in the beginning. I decided to do another rough roto for the wall, as it is in the foreground and doesn't really need to be tracked. My track should just track the reflections and windows, which you can see in the reflection. My next track was also not working, because the card seemed to slide in the end.
After a few adjustments and some more tracks, I managed to get a good track and the card was finally sticking and staying in place
02 DECENT PAINTWORK
The card looked fine and I started to paint first with the framhold paint tecnique and than I tried the UV projections. Altough I had a cleanplate, it was not useful: in the cleanplate you can see the reflection of the camera man. Dealing with the reflections is the hardest part in this shot. It seems easy, but in the reflection you still can see backbags and other stuff, it's not an even, unicolored texture that has to be restored.
First I thought using one card for the window and the mirror, but it didn't work out! Next thing I tried is to split it up and place two cards: one for the window and one for the mirror. To make it as accurate as possible, I had to find a way to create a good edge between the mirror and the wall behind it.
03 FINDING THE RIGHT TECHNIQUE
Trying to do a good paint job didn't work out quite well, using UV projections was the better solution, but still not a good enough result in my opinion. I came up with another solution to restore the edges and the wheels from the mirror: Creating a roto shape around the area which I wanted to retain, was the best solution and gave me a pretty good result!
04 BRING BACK DETAILS
Bringing back the bars, one in front of the mirror and the others in the reflection of the mirror was another problem I had to face. A normal roto node combined with a blur node was not working, due to the shadow of the cameraman. After not getting the desired result, I UV projected the bar in front of the mirror and made a separate UV projection for the reflection of the bars. As I was not the only one working on this shot, someone else used a gizmo called Pxf_filler from Pixelfudger. This gizmo takes pixels from the left and right side of your roto. In theory using that after a roto would cover the dark shadows from the cameraman. In my case this wasn't working because the arm of the dancer was in the way (the PxF_Filler would take the texture of her skin to fill in the edges, which is not useful). In the end the UV projections were the best option!