We have a brand new look! Take a tour with us and explore the latest updates on Adobe Support Community.
I cannot for the life of me figure out what I am doing wrong. I have had really good results with rotobrush but now it simply just seems to not be working well at all, and I would really appreciate some help!
I am trying to select a golden fork on a black background. In the original uncolored image there is quite a bit of contrast between the fork and the background, yet it is proving very difficult to select even the base frame properly. I read some advice on this forum about adding contrast to the clip before the rotobrush, and so I did that. Still I seem to be getting the exact same results.
After painstakingly setting down the base frame on what feels like should be an easy select, the tracking loses the mask almost immediately. I have attache screenshots below showing some of my results. These selection results are only two frames apart.
Closeup starting Frame:
After Tracking Two Frames:
This last frame (after scanning) is also what my selection looks like when I make a first attempt at selection, to me it seems so odd that it would lose it this much!
Here is the image without contrast added and after selecting the tines with the rotobrush. Even though i drew up the whole tine, it only selected the lower half and erroneously selected the inbetween:
Here we go after a first attempt at removing the space in between both tines:
To get the selection good on this fork takes a solid 5-10 minutes and lots of retries and going over the same piece. Then the scanning/tracking loses the selection almost immediately.
Even weirder is that the additional contrast does not seem to help either problem.
Any advice would be greatly appreciated! Thanks very much!
Sorry one more important point: I used "BEST" quality and had the same results with the scanning. I screenshotted these settings after having tried to use the "STANDARD" scanning as a last resort, the results were worse but the "BEST" were really not much better, the other screenshots were all done with the settings on "BEST"
And another example of the difficulties I'm dealing with here is this:
I am selecting this part for the first time:
The result: no selection:
Huh? You're simply using the wrong tool. This is clearly a job for conventional masks. The time you invested in fixing the RB stuff could already have given you some usable frames with normal masks. Sure, manual masking is tedious, but even in this day and age sometimes unavoidable...
Thanks very much for the advice! Yes I definitely agree and I have taken this approach for this specific problem. More than anything I was trying to understand what it is about the rotobrush that prevents it from doing a good job in this particular situation. The goal being more that I know what to expect when I approach similar challenges in the future. For this particular problem I am using a combination of a mask to select that area and clean matte just as Rick has suggested below.
It would take about a minute to turn that black background into a very effective and clean matte using extract.
You could just add curves and extract to a copy of the original footage, then set that as an Alpha Track Matte for the original and get this in under 2 minutes.
Rotoscope and Rotobrush should be your last tool for creating transparency. Procedural mattes using colors, color channels, or luminance values are much preferred and a lot faster. Properly shot footage can give you nearly perfect results. Even imperfect footage can be helped a great deal by using a combination of procedural mattes, masks, and rotobrush.
Thanks so much for taking the time and effort to respond and give me some advice on this! I definitely have commited to using a combination of tools to get a matte out of this as you have suggested. And I fully agree that sometimes it's not what should work but was does work which is the best solution.
I just could not understand how this rotobrush tool worked instantly and perfectly on some areas of my footage with so much less contrast but would not work at all on these forks! It was bewildering to me! For example, look at this plate, it moves into the frame very quickly and with a great deal of motion blur. It took me about 1 minute to create the base layer and the rotobrush did the rest of the work to perfection. I had nothing else to do once I had finished.
I guess I am just trying to understand what property of the rotobrush prevents it from working well on the forks but allows it to work on the plates, so that I know better for the future what I can plan to use it for. Do you think it might be accurate to guess that the "smallness" of the fork tines pose a greater challenge to Rotobrush? Could it be that the large area of the plate being associated to the edges helps the rotobrush figure out what to select?
Rotobrush works on more than just color and luminance values. It is trying to figure out what is moving and what should stay together. There is simply not enough data in the image or the fork. Rotobrush was just using luminance, maybe the fork shot would work better.
Your bowl shot isn't that easy either:
I'm up to 14 foreground strokes and 6 background strokes and it is not anywhere close enough for a good clean matte.
Rotobrush is always my next to the last masking option. Roto by hand is the last option. Sometimes, plain old rotoscoping with masks is the least amount of work, sometimes you need to combine all masking tools to create a good matte. I'm just wrapping up a composite that uses stabilized power pin tracking to stop movement, a track matte from a solid with an animated mask to fix one part of the stabilized shot, extract to fix another part of the shot, and rotobrush to fix the actors hair on a tiny part of the shot. Trying to create the entire matte using only one of these tools would not have produced anything close to an acceptable result.
Wow that's actually quite interesting. It may be easier with my bowl shot on my computer because the source image is 4k so the lines are more defined. I actually did a fairly rough roto and then just crudely painted over the edges with the Refine Edge tool and it worked like an absolute charm. That being said, this new difficulty with the fork made me realize that as you say, the tool is using more than just luminance to define the edges. Thanks very much for your input and alsso letting me know that the pros used a mixed bag of methods to achieve good results. That is what I was doing and it felt a bit janky- but I suppose that is how it goes! Thanks again 🙂