Line rasteriation issues for dense plots about implot HOT 17 CLOSED

Any comments about this issue ? It'd be nice to know your ideas on further developments in this area (if any)

from implot.

Hi @sergeyn , sorry for the slow reply and not addressing open PRs. I've been busy with other work and hope I can get to them soon.

Regarding this issue, I think this falls under the need to (optionally) downsample user data. For example, see the documentation for pyqtgraph:

from implot.

But I'm not sure how exactly does this addresses the issue. The problem is that points at the same place from neighbor line segments don't end up at the same spot. With dense rendering like I do problem just becomes obvious, it will not go away if you perform a reduction, unless you reduce everything to one line of course.
I mean that something like this: https://github.com/mattdesl/extrude-polyline needs to get implemented to properly draw connected segments.

from implot.

Well, downsampling reduces the chance that you have many points hitting a single pixel, I suppose.

Regarding connected segments: If I'm not mistaken, this is what ImGui's AddPolyline does by default (I can't remember for sure, perhaps I'm wrong). Can you swap out ImPlot's implementation with a call to this function and see if it improves?

from implot.

Will try , thanks. I'm honestly not that familiar imgui internals yet. Another way I am thinking about is to push actual line strips to gpu. This would reduce amount of data generated and you get thin connected lines without much effort.
The default line rendering you have is for performance reasons, correct ? I could further significantly optimize line rendering but I don't feel like investing time into a method which doesn't produce accurate result.

from implot.

I agree that we could get optimizations by bypassing ImGui and going straight to the GPU, but I'm hesitant to support that right now since it would be a massive undertaking to handle all the potential backends (we'd essentially have to provide an implementation for every example backend ImGui does).

from implot.

@sergeyn , could you post some screenshots of what exactly is happening here. I'm curious.

from implot.

from implot.

@sergeyn , I'm testing your PR. Can you provide the code for the plot above?

from implot.

         [](void*, int idx) {
            float r = 0.9f; // outer radius
            float a = 0; // inner radius
            float b = 0.05f; // incerement per rev
            float n = (r - a) / b; // number  of revolutions
            double th = 2 * n * M_PI; // angle
            float Th = float(th * idx / (1000000 - 1));
            return ImVec2(0.5f+(a + b*Th / (2.0f * (float) M_PI))*cosf(Th), 0.5f + (a + b*Th / (2.0f * (float)M_PI))*sinf(Th));            
         }, nullptr, 1000000);```

Note that when plot completely gets culled away, you may see garbage on screen (same as you have in your master currently), which is fixed by [this](https://github.com/ocornut/imgui/pull/3232)  PR. It's not a problem of plot rendering.

from implot.

Fixed in #41

from implot.

May I ask how exactly is this fixed ?
As you can see on the image - here almost every other pixel is missing.

from implot.

Apologies, I misremembered the issue. I thought this one was related to the spiral truncating.

Yes, this does seem to still be an issue. As I said before, I think the best solution for this is to down sample the data with the user's permission.

from implot.

from implot.

Out of curiosity, what sort of data are you trying to plot where this is a real issue?

from implot.

time series with possibility for extreme zoom-outs. I simple want them to look nice.

from implot.

I'm closing this issue because:

1. We are never going to support mitered edges...the cost is just too high
2. Downsampling, as you said, should NOT be default, and it's actually quite simple to implement using the existing offsetting and striding features. An example of this can be seen in the new Time Formatting demo, where 31,622,400 points are dynamically downsampled and plotted:

from implot.