-
-
Notifications
You must be signed in to change notification settings - Fork 21.4k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Improve visual feedback when using the motion vectors debug view option #80723
Conversation
Although this is a bit subjective, I believe the motion vectors should be in the opposite direction as the current video. The rationale is that this is a debugging feature and when you're debugging MV (motion vectors) bugs, you think in terms of where the current pixel used to be before, so the arrow should be pointed to where the pixel was. Otherwise you're debugging a pixel from current frame with an arrow that indicates "look up", but there is nothing to look up. You need to look down, which is where that pixel used to be and that's where the information was. In fact that's where the motion vector buffer is pointing to (the raw data numbers). Without the debugging mindset, it sounds tempting to use the direction in the current video: it feels more intuitive that the arrow points to where the object is traveling to, as if it were some sort of gameplay feature. However this is not a gameplay view, it is a debugging one. With the current arrows in the video, one has to mentally revert the direction which can be very exhausting and can lead to errors. Not everyone has the same thought process though, so it can be a bit subjective. |
@darksylinc To be fair, Dario implemented the vectors pointing the other way originally and I asked him to switch the direction as I found it more intuitive the way he has it in the OP. I think seeing as both of you agree that the other direction is more intuitive, it makes sense to switch back |
It's worth mentioning that going by Matias' argument of seeing the raw values, it should be clarified that these are the real raw values of the motion vectors. It did strike me a bit weird as usually the method I've seen (and the one FSR2 expects as well as I had to use a negative multiplier) is the way Matias described: the value that points to the previous pixel. Sure enough if you look at taa_resolve.glsl, it becomes apparent it compensates for this direction in the same way I was originally doing as well: // Get reprojected uv
vec2 uv_reprojected = uv - velocity; And sure enough that is because motion_vector = position_uv - prev_position_uv; Which results in a vector pointing the opposite direction. Maybe it's worth evaluating to flip the sign of the value being stored and fixing it on the TAA pass instead. I feel that'd fit the convention of motion vectors that I've usually seen in most engines. |
Both directions have their merits, maybe expose this to editor settings? |
0b5fdd9
to
c82aa95
Compare
Replaces the current method of showing the raw values of the motion vectors buffer to display a grid of lines instead with a new shader.
c82aa95
to
e7d3a7c
Compare
Added the suggested changes and a red color for NaN values if they're detected as per Matias' suggestion. This should be ready for reviewing. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks great!
Thanks! |
I don't think we need to cherry-pick this anyway, it was just to make my life easier while debugging. Motion vector debug views aren't really used by the wider public yet. It can wait for 4.2 with FSR2. |
Replaces the current method of showing the raw values of the motion vectors buffer to display a grid of lines instead with a new shader. This should help make debugging motion vectors much easier.
This currently replaces the current method of viewing motion vectors. It should be decided whether to keep the old one as an option or not. Personally I don't see much use of it as if I want the raw values, I'd much rather inspect the texture on RenderDoc than look at RGB values that can't properly display the low range they usually have.
mvs.mp4
Side note: @clayjohn and I couldn't agree on what direction the vectors should point, so we'll let the people decide. :)