imagej/imagej-ops

ImageJ Ops: "Write once, run anywhere" image processing

Stefan Helfrich

@stelfrich

@ctrueden It lives in https://github.com/imagej/imagej-benchmarks

Gabriel Selzer

@gselzer

@ctrueden Question: should classes such as Arity0-OT have a compute(O out) method?

Gabriel Selzer

@gselzer

I guess maybe not, seeing as how if you had an O out in mind you would just pass that to the Op builder, not the O?

Gabriel Selzer

@gselzer

@ctrueden inplaces are now supported in the OpBuilders. I added computer/inplacen methods to the OT op builders, seeing as how I didn't see any reason not to add that functionality

Curtis Rueden

@ctrueden

@gselzer Thanks. And yeah, the builders don't have apply/compute/mutate methods that accept args. Those should have been given already. If you want that, request the function/computer/inplace and use it directly.

Gabriel Selzer

@gselzer

Yeah, that was what I was assuming we would do

Gabriel Selzer

@gselzer

@ctrueden I still feel like the inplace support for OpBuilders is still a little wonky:

        final double[] actual1 = {1, 2, 3};
        final double[] actual2 = {1, 2, 3};
        final Inplaces.Arity2_1<double[], double[]> castToInt = //
            name("test.mulArrays").input(actual2).output(actual1).inplace1();
        castToInt.mutate(actual1, actual2);

Curtis Rueden

@ctrueden

@gselzer Shouldn't the input call be input(actual1, actual2)?

Gabriel Selzer

@gselzer

Ideally, yes, however if you did that here you would get an Arity2_IV_OU, which we don't want with an Inplaces.Arity2_1

So we would (will) have to restructure

Curtis Rueden

@ctrueden

which we don't want

Why not...?

I think the Arity2_IV_OV needs to support calling inplace1() to match an Inplaces.Arity2_1, and inplace2() to match an Inplaces.Arity2_2. No?

Gabriel Selzer

@gselzer

Hmm, I suppose we could, actually, I just didn't do it that way earlier since then the type parameters were weird. But I do suppose that conceptually this would probably be a sound change.

Although this could cause some type parameter shenanigans. We just have to trust that the users are smart enough to have the io argument in inputbe the same type as the argument to output

Gabriel Selzer

@gselzer

Interestingly enough, if we do this then we really don't care about the output at all. So then Arity2_IV_OU/Arity2_IT_OU can have inplace1/2 methods

Gabriel Selzer

@gselzer

Although then I suppose this brings up the question: do we want inplace1/2 methods in the OpBuilders where we have declared output types/values?

Gabriel Selzer

@gselzer

I would argue no, seeing as how if someone calls output(<some type/value>), that type/value will not be used if the user then calls .inplace()/.mutate(), so we don't want to be misleading by allowing that.

Curtis Rueden

@ctrueden

We just have to trust that the users are smart enough to have the io argument in inputbe the same type as the argument to output

You can check. And if it out is not the same reference as in1 then calling inplace1 can throw IllegalStateException.

Arity2_IV_OU/Arity2_IT_OU can have inplace1/2 methods

Sure, I think that would be fine.

do we want inplace1/2 methods in the OpBuilders where we have declared output types/values?

If you do the reference checking I mentioned, then it makes sense too, I think. But I don't feel strongly.

Jan Eglinger

@imagejan

The complex plane of Gaussian prime numbers looks beautiful: see this forum topic.
Can we generate this using ops? I imagine something like this:

ops.run("equation", img, "isPrime(p[0]*p[0]+p[1]*p[1])”)

… but we would likely need a lambda expression to define isPrime, right? Do we have to go Java, or is this possible in Groovy,Python,etc. ?

Curtis Rueden

@ctrueden

@imagejan IIRC, @bnorthan added support for passing Functions to the equation op?

This one I guess:

    (IterableInterval out) =
    net.imagej.ops.image.equation.DefaultXYEquation(
        IterableInterval out,
        DoubleBinaryOperator in)

Curtis Rueden

@ctrueden

#@ OpService ops
#@ int width
#@ int height
img = ops.run("create.img", [width, height])
java.util.function.DoubleBinaryOperator gaussPrime = {
    x, y -> new BigInteger((long) (x*x+y*y)).isProbablePrime(1) ? 1.0 as double : 0.0 as double
}
ops.run("image.equation", img, gaussPrime)
img

Jan Eglinger

@imagejan

Thanks, @ctrueden, that's great!

Gabriel Selzer

@gselzer

@ctrueden OpBuilder is now auto-generated!

Will work on Functions / Inplaces now, then will finish up with the tests

Gabriel Selzer

@gselzer

@ctrueden Code gen is functionally complete, will clean up with the little time left. What arity do we want to generate to?

Curtis Rueden

@ctrueden

What were we at before? 13 or so?

Maybe 16 for now?

Gabriel Selzer

@gselzer

Woah, I think we maxed out at 11

I can do 16 if you want

Curtis Rueden

@ctrueden

I like powers of 2. :-)

But 11 or 12 is also fine for now.

It's worth fine-tuning the templates a bit before we generate everything, since every time we tune them it will result in a largish git commit.

Gabriel Selzer

@gselzer

Looking at this hurts my brain :grin:

public interface Arity16<T, U, V, W, X, Y, Z, A, B, C, D, E, F, G, H, I>{

Curtis Rueden

@ctrueden

I favor using numbered parameters over letters.

Otherwise we'll sometime run into the Y26 bug.

Gabriel Selzer

@gselzer

Sure, I just used letters for Consumers because this was the convention we used before. I can change to numbers if you would prefer.

Gabriel Selzer

@gselzer

Did we come to a conclusion about I1 vs I for the unary ops?

Curtis Rueden

@ctrueden

I favor whatever makes the Java code most usable by programmers. The vast majority of people will never see the template, so it is OK if it's a bit more complicated. Whereas more people will look at the generated code and especially the API, which includes the generic parameter names.

Therefore, I lean toward I instead of I1.

Gabriel Selzer

@gselzer

Okay, sounds good

Gabriel Selzer

@gselzer

@ctrueden is there a good way in groovy to combine two streams?

Hmm, maybe I should say, something like zipping two streams

Curtis Rueden

@ctrueden

The same way as in Java?

I am no Groovy expert. Google will help you better.

Gabriel Selzer

@gselzer

Ah, never mind, found it :grin:

Curtis Rueden

@ctrueden

:+1:

Gabriel Selzer

@gselzer

Was not looking in the right place yesterday

Where communities thrive

People

Repo info

Activity