Check out the pdf, pretty impressive results:

https://www.ri.cmu.edu/wp-content/uploads/2019/01/Li-Mengtia...

BTW, if the authors find these comments, they should definitely advertise the new game they create to collect more data (once it's released) to art-centered communities, particularly comics. Many people who do line art practice tracing frequently to improve their technique.

Hmmm. The examples are underwhelming compared to commercial examples of these types of filters.

iOS 12 added filters which are only available in Messages or the Clips app (not the camera app) that produce amazing results in realtime. Try the "ink" version. https://www.cultofmac.com/583400/how-to-use-ios-12-comic-boo...

For even more extreme versions, there are tons of filters, plugins etc for Photoshop and other paint programs. Here's one from 2013: http://www.tipsquirrel.com/the-graphic-novel-filter-photosho...

Maybe I'm missing something, but why is it that academics manage to publish stuff like this that is so far from the state of the art in the commercial world?

> Maybe I'm missing something

This may be the case. The filters you cited are not countour filters. Having the option of pixel matching entire regions by color makes any image look better, regardless of how many or how few contours you match.

FYI it seems like this filter relies on Apples’s latest depth detection hardware not just an image. I have a pre X phone and that option doesn’t come up for me.

Their code make actual outlines, it doesn’t merely detect edges and applies a transform to those pixels. Basically they can make inks automaticelly, if poorly.

It's 2019. Can we please stop using the Lenna Playboy centerfold picture in every other graphics paper?

Perhaps in the 70s there was actually a shortage of useful reference images and it had some value -- I wouldn't know -- but at this point it's just creepy.

I don't understand this level of hostility against nudity. A remarkable portion of renaissance art involves nudity as well as display of genital parts (which isn't even the case with Lenna BTW). If anything, there's this thing called freedom of expression.

I personally have no problem with someone using the face of Lenna or Fabio or Super Mario, or Michelangelo's David, or whomever in a paper. It's a face, and it's their paper. No, it's not OK for you to shame them into not doing that just because it doesn't conform to your particular culture/sense of ethics.

People imposing their ethics, typically rooted in a particular religion, onto others isn't new, but bashing use of Lenna's face is in particular hip nowadays. No, I don't agree that her face from that photo is somehow tainted because the original photo involved nudity, or that she is tainted/irredeemable sinner because she appeared on Playboy. I'm an atheist, and neither I sympathize with this sense of ethics commonly found in Abrahamic religions, nor I appreciate it being imposed onto others. Can't people be more open minded and tolerating to differences? "It's 2019."

And regarding the feminist propaganda surrounding this, claiming Lena is offending female students in computer science is tantamount to claiming that renaissance art is offending female art students.

From https://www.cmc.edu/news/every-picture-tells-a-story:

> And then maybe the next woman to take a computer science class won’t have to hear the snickering. Maybe, she’ll feel just a little more welcome among the mathematicians, scientists and engineers.

You can hear snickering in a biology class on sexual reproduction as well. And there is indeed a push for elimination of education on sex in various countries including the US due to pressure from religious conservatives. That wouldn't be the solution in this context, because it doesn't address the problem of sexual harassment. That kind of people don't need Lenna to do their "snickering", and Lenna isn't the reason that they're "snickering". That's just an excuse to justify a political/religious agenda.

Imagine this: there's a group of grad students on a field trip, one is a girl from Colombia. A guy starts listening to Hips Don't Lie loudly whenever she comes near him. Should we ban Shakira on campuses because she can be used as an accessory for sexual harassment? And do you believe it would solve the problem, or would those people just move onto something else and continue doing their sexual harassment?

Introducing a mandatory lecture on sexual harassment (for all students, faculty and staff), improving the on campus response mechanism and truly penalizing sexual harassment, on the other hand, would be an example of something that would actually help address the problem.

Some people are just not happy unless they are unhappy. Or outraged. Or both.

My initial reaction to this was "what's wrong with it, it's a totally SFW shot of someone from the shoulder-up, mildly suggestive at its very worst?", then I challenged that thought a little. I can see why it would be an issue to sit as a female student in a classroom in a heavily male-dominated field and see that we're using Playboy centerfolds as a source of data when perfectly good data of the same type is functionally limitless. It sure doesn't do much to dispel the notion that computer science academics is a boys club, does it?

Many physics, engineering and maths departments are still perceived as "boys club" as well. All without Lenna. Food for thought.

What's the issue with the image? In the form used it's not even remotely explicit, or really even suggestive. Do you have some moral objection to consenting pornography? On what grounds? People just shouldn't be sexual? What business is it of yours what Lenna chose to do with her body?

Agreed, it's 2019 I too would rather have the computer vision community use a different image instead Lenna, perhaps the proposed Fabio Lanzoni[1] or the Kodak image suite[2].

[1] https://www.cmc.edu/news/every-picture-tells-a-story

[2] http://r0k.us/graphics/kodak/

I like the idea of changing it, but neither of those options seem good. Fabio is just a response to Lenna - it seems like we can do better. And those Kodak images are really surprisingly bad. Consistently drab colors, blown highlights, and really badly composed pictures of dated subjects. Is there something I'm missing that makes those images important?

Fabio is fine, and he is proof that there's nothing wrong with Lenna either. Papers with algorithms relevant to photographs of humans should probably include both images, as well as a few more to include some racial diversity.

Images of humans are important to humans. If dogs were scientists, they'd probably devote lots of study to the scents of other dogs.

The image may be fine if we strip it of the context, but the world doesn't work like that. The image is famously known to be cut out from a larger nude photo, which was chosen for its sexual appeal, and published in a magazine of sexual content. That is not an appropriate choice for most academic publications.

If you want to use it to demo your nudity detection algorithm, feel free.

A noted judicial maxim, "I know it when I see it", has been criticized as trivial. Here you prove that not to be the case. You have a serious disagreement with the maxim, in that the purely visual properties of this image are not what make it objectionable to you. Instead you dislike the history of the image. But its nearly five decades of use in this field are the historical reason it is used today. You're advocating for your historical judgment over the different historical judgment of scholars in the field.

That could be reasonable. For instance, the fact that statues commemorating the Confederacy have some historical relevance doesn't make up for the fact that they mostly celebrate the awful institution of slavery and a society that was built upon that. We therefore disregard the "historians" who advocate for their continued intrusion into our public spaces.

However, those statues are quite explicit in their celebration of that odious time. This image is simply a beautiful image of a beautiful human. And its history is far from odious. Even the larger image is hardly prurient: secondary sexual characteristics are viewed obliquely. Most women have no problem with that. (Instead they object to the continued vilification of breasts in the form of breastfeeding bans etc., which vilification you continue here.) Lenna herself certainly has no problem with the image: she has publicly stated she values her appearance in the best-selling Playboy of all time, and its subsequent second life in academia. There is nothing wrong with an image of an unclothed human body, and this image is tamer than many such images.

Few women appreciate your advocacy here. Unlike you they don't assume that everyone is either a Puritan or a junior high schoolboy. I can think of few things less interesting or important than a "nudity detection algorithm".

It's not clear why the objections lodged against the original image could not also be lodged against Kodak's "kodim04" image. Both are images generated in the 1970s that portray smiling attractive young women of Scandinavian descent with bare shoulders.

[EDIT:] take a look and tell me I'm wrong: http://r0k.us/graphics/kodak/kodim04.html

Because the original image was cropped from pornography.

Quality doesn't seem much better than the comparison algorithms in the paper.

Which images are you looking at? The ones I've seen appear dramatically different and better to me(but I'm also a hobbyist artist and sketcher).

E.g. on page 7 of the paper where the head is cut off in the "ours" picture and less so in the "HED" picture, or the helicopter is similar or worse in quality to the "gPb" picture.

Sure, it's not 100%. They included those images to illustrate some challenges that they ran into (negative space and low value boundaries, not surprising given their relatively small data set).

But the ones that they open with are dramatically better than HED.

Hmm, I'm getting downvoted, but even the opening ones look arguable to me, e.g. in the hotel room on the first page the lamp shade and cushion near the sofa don't look great. Some of the HED artifacts could be improved with some post processing I imagine (as they do later). The dogs are significantly better on page 6. Looks to me like there are some things they are doing better, but still a way to go. Difficult to tell with the small number of sample images.

I didn't downvote, but I still see a significant difference between HED and the author's algorithm.

But you're correct, it is difficult to make a confident judgment given the small number of sample images.

Likewise. The the new technique produces dramatically more distinct edges, closing gaps where HED et al. taper off and finding meaningful interior edges that others omit without also pulling in noise edges.