Нам очень жаль, но эта информация доступна только на английском языке. Читать машинный перевод текста
Everyone claims they have cutting-edge AI and are the best there is. So how do you separate the wheat from the chaff?
The first thing to look for is obviously the background removal itself: can the tool actually separate out the background from your particular photos? And beyond that immediate task, does it serve your ancillary needs as well? Here are seven things to keep in mind when you make your pick:
Here you primarily want to look for gross errors in the result: blobs sticking out where they shouldn't, pieces of foreground getting cut off, and similar.
The quality of the clip depends primarily on the Deep Learning architecture used and the examples that were available for training.
Most background removers base their network on a published paper, use publicly available datasets consisting of at best tens of thousands of images, add some data augmentation and call it a day. That approach leads to mediocre results with frequent errors when subjected to real world images.
In contrast, Clipping Magic uses millions and millions of images uploaded by real users. This means that we have orders of magnitude more data than the competition. Also, our data matches the broad range of images that people want to remove backgrounds from, not the narrow niches that some academics conveniently published papers on. This gives our AI a breadth that the competition could only dream of.
Here you want to look for three things:
In an effort to keep noise out many background removers do a pretty tight crop.
This can lead to an unfortunate loss of "liveliness" in the hair, where too many of the fine strands are lost.
It can be hard to model the colors in the partially transparent areas, and they easily end up looking flat and without the natural texture typically present in hair.
This can lead to dull or flat looking hair strands.
The flip side of the previous point: sometimes the background colors gets pulled into the foreground, leading to "holy hair halos" where it looks like your subject is backlit by a background-colored light.
Hair is a challenge for background removers primarily because it's hard to get good training data for it. This is both because it's much more work to do a good separation of hair, so each training example is more expensive, and because you need more training data for this more nuanced task. So most efforts resort to using a small number of well clipped foregrounds and then use data augmentation - a fancy word for "copy this foreground onto a bunch of backgrounds" - to get enough volume.
Needless to say that method has its limitations.
Clipping Magic has again taken a different approach. By scaling a customized machine learning method to run >10x faster that it normally would, we've been able to provide best-in-class real time hair background removal that is fully editable.
This means we not only remove the background from the hair fully automatically, we also show you what part of the image is partially transparent, and allow you to edit that mask to suit your needs.
It's easy to get a cutout of a perfectly staged shot taken against a white-ish background to look good when you put it over a perfect white background.
It's trickier when a shot was taken against a busy background, or when you want to put the result over a background with a completely different color.
Here are a couple of things that set good edges apart from bad ones:
At the edge of an object, the background color inevitably seeps in to some extent.
A bad background remover will just include this in the result.
A good background remover models the colors at the edge and actively separates out the foreground from the background color.
This is most visible when you put the result over a new background with a very different color than the original background. If your result all of a sudden gets a little glowing outline in the original background color, then you have a halo situation on your hands.
An object's edges can vary greatly in how sharp they appear. Some photos were taken with extreme focus blur (bokeh), others were effectively taken with a pinhole camera making them super-crisp. As usual, most are somewhere in between. The edges also usually have a little bit of texture in them. They're not entirely flat or uniform in color.
Really bad background removers ignore all of this and produce a result with a hard outline. Those results look strangely artificial in a hard-to-put-your-finger-on-it kind of way.
Bad background removers do feather, but ignore the grit. This creates a passable, but less-than-ideal result.
Good background removers apply appropriate feathering and include the grit for a great, natural looking result.
The dirty little secret in background removal is that no solution will ever be 100% correct on all images. So for some of your images you will either have to make do with an obviously wrong result, or you will have to edit it.
Most background removers offer at best crude pixel-level correction capabilities, reminiscent of using MS Paint back in the day to erase an image's background.
Clipping Magic on the other hand offers you a Smart Editor powered by an algorithm that fuses classical machine learning with deep learning, giving you the best of both worlds. You can inspect your automatically produced result in a full pan-and-zoom interface (like Google Maps), and when you apply corrections, the algorithm makes them go much further than simple pixel edits.
This makes editing an actually realistic option. And it enables you to succeed on 100% of images - no need for a fallback solution!
While it certainly happens, it's rare that literally all you needed was to remove the background from an image. Typically you also want to crop it, mount it on a white background, spruce up the colors, and maybe apply a shadow or reflection effect.
Most background removers leave you stranded, limiting their post-processing to switching the background color.
Clipping Magic on the other hand strives to be your complete image processing solution. Our goal is to enable you to go straight from camera to Clipping Magic to publishing ready result.
To that end we offer a comprehensive portfolio of post-processing tools and effects:
Web optimized output so your images load quickly.
Most background removers are exclusively focused on removing the background from a single image. They also typically give you very limited online preview options. You usually have to download a small preview and then open that in a separate viewer to actually see the result in all its glory. This makes them a bit awkward even for sporadic use and downright painful when you have a large stack of images to plow through.
In contrast, Clipping Magic has been built from the ground up to leverage the full power of your browser and to minimize repetitive actions.
This is embodied in our Smart Editor, extensive configurable defaults, and the minimized number of clicks it takes to clip each image. For most images it's literally drop on page, click download, done. But when you want more, pan-and-zoom review, touchup, and all the other tools are right there. And our bulk clipping mode lets you zip through even the largest catalogs, producing consistent publishing-ready results.
No other background remover has gone to such lengths to fully streamline and supercharge your workflow.
While price shouldn't be the first thing on your list of considerations, it is always and inevitably a consideration.
Some of our competitors claim to be free. This inevitably comes with some fine print and caveats.
In general, the really free options tend to offer little in terms of quality, capabilities and workflow, making you trade off your time spent using the tool and the consequences of a mediocre result for any cash outlays.
The higher quality options sometimes offer one or two credits on signup, but still typically have very limited capabilities and a higher sticker price than Clipping Magic once you're actually using them.
At the end of the day we think you'll find that you in reality pay less and get more with Clipping Magic than with the competition.
So if you care about quality and value your time at all, then you'll find Clipping Magic to be a solid choice.
Thank you for reading! We hope this has given you a good foundation for picking a background remover.
In short, be sure to look beyond the AI and Deep Learning hype and pick a tool that serves your overall needs well.
Вы загрузили изображение, размер которого превышает ваш активный лимит, поэтому нам пришлось уменьшить его до 4 мегапикселей. Вы можете настроить лимит размера изображения или включить предварительную обрезку в диалоговом окне Параметры ввода.