Understanding Android’s vector image format: VectorDrawable
Android devices come in all sizes, shapes and screen densities. That’s why I’m a huge fan of using resolution independent, vector assets. But what exactly are they? What are their benefits? What are the costs? When should I use them? How do you create and use them? In this series of posts I’d like to explore these questions and explain why I think that the vast majority of the assets in your apps should be vectors, and how to get the most out of them.
Raster vs Vector
Most image formats (png, jpeg, bmp, gif, webp etc) are raster which means they describe the image as a fixed grid of pixels. As such they’re defined at a particular resolution and don’t understand anything about their contents, just the color of each pixel. Vector graphics however describe the image as a series of shapes defined over an abstract canvas size.
Vector assets have 3 main benefits, they are:
Vector images resize gracefully; because they describe the image over an abstract canvas size you can scale this canvas up or down and then redraw the image at that size. Raster assets however can deteriorate when you resize them. Scaling raster assets down tends to be ok (as you’re discarding information) but scaling them up leads to artifacts like fuzziness or banding, because they have to interpolate the missing pixels.
This is why on Android we need to provide multiple versions of each raster asset for different density screens:
Android picks the closest larger density and scales it down (if needed). With the trend for devices with ever higher density screens, app makers must keep creating, including and shipping ever larger versions of the same assets. Note that many modern devices don’t sit on exact density buckets (e.g. the Pixel 3 XL is 552dpi, somewhere between xxhdpi & xxxhdpi) so assets will often be scaled.
Because vector assets resize gracefully, you can include a single asset, safe in the knowledge that it will work on any and all screen densities.
Vector assets are generally* more compact than raster assets both because you only need to include a single version, and because they compresses well.
For example here’s a change from the Google I/O app where we switched a number of icons from raster PNGs to vectors and saved 482KB. While this might not sound like much, this was just for small iconography; larger images (such as illustrations) would have larger savings.
This illustration for example from the on-boarding flow of a previous year’s I/O app for example:
We could not replace this with a
VectorDrawable as gradients were not supported widely at that time (spoiler: they are now!) so we had to ship a raster version 😔. If we had been able to use a vector, this would have been 30% the size for a better result:
- Raster: Download Size = 53.9KB (Raw file size = 54.8KB)
- Vector: Download Size = 3.7KB (Raw file size = 15.8KB)
Note that while Android App Bundle’s density configuration splits bring similar benefits by only delivering the required density assets to the device, a
VectorDrawablewill generally still be smaller and also removes the need to keep creating ever larger raster assets.
As vector images describe their contents rather than ‘flattening’ them down to pixels, they open the door to interesting new possibilities like animation, interactivity or dynamic theming. More on this in future posts.
Vectors do have some drawbacks that need to be considered:
As previously stated, vector assets describe their contents, therefore they need to be inflated and drawn before use.
There are two steps to this:
- Inflation. Your vector file has to be read and parsed into a
VectorDrawablemodeling the the paths, groups etc you declare.
- Drawing. These model objects then have to be drawn by executing
Both of these steps are proportional to the complexity of the vector and the type of operations you perform. If you use very intricate shapes, it will take longer to parse this into a
Path. Similarly, more drawing operations will take longer to perform (and some are more expensive e.g. clip operations). We’ll revisit this in a future post in this series on profiling these costs.
For static vectors, the drawing stage only needs to be performed once and can then be cached to a
Bitmap. Animated vectors, can’t make this optimization as their properties necessarily change requiring re-drawing.
Compare this to raster assets like PNGs which only need to decode the file’s contents, something which has been highly optimized over time.
This is the essential tradeoff of raster vs vector. Vectors provide the aforementioned benefits but at the cost of being more expensive to render. In Android’s early days, devices were less powerful and screen densities differed little. Today, Android devices are more powerful and come in a huge variety of screen densities. This is why I believe it is time for all apps to move to vector assets.
Due to the nature of the format, vectors are great at describing some assets like simple icons etc. They’re terrible at encoding photographic type images where it’s harder to describe their contents as a series of shapes and it would likely be a lot more efficient to use a raster format (like webp). This is of course a spectrum, depending upon the complexity of your asset.
No design tooling (that I know of) creates
VectorDrawables directly which means that there is a conversion step from other formats. This can complicate the workflow between designers and developers. We’ll go into this topic in depth in a future post.
Why not SVG?
SVG does however include a path spec which defines how to describe and draw shapes. With this API you can express most vector shapes. This is essentially what Android supports: SVG’s path spec (plus a few additions).
Additionally, by defining its own format,
VectorDrawable can integrate with Android platform features. For example working with the Android resource system to reference
@strings, working with theme attributes or
AnimatedVectorDrawable using standard
VectorDrawable supports SVGs path spec, allowing you to specify one or many shapes to be drawn. It’s authored as an XML document which looks like this:
Note that you need to specify the asset’s intrinsic size, which is the size it would be if you set it in a
ImageView. The second
viewport sizes define the virtual canvas, or coordinate space all subsequent drawing commands are defined in. The intrinsic and viewport dimensions can differ (but should be in the same ratio)—you could define your vectors in a 1*1 canvas if you really want.
<vector> element contains one or many
<path> elements. They can be named (for later reference e.g. animation) but crucially must specify a
pathData element which describes the shape. This cryptic looking string can be thought of as a series of commands controlling a pen on a virtual canvas:
The above commands move the virtual pen, then draw a line to another point, lift and move the pen, then draw another line. With just the 4 most common commands we can describe pretty much any shape (there are more commands see the spec):
C(cubic bezier) curve to
Zclose (line to first point)
(Upper case commands use absolute coordinates & lowercase use relative)
You might wonder if you need to care about this level of detail — don’t you just get these from SVG files? While you don’t need to be able to read a path and understand what it will draw, having a basic understanding of what a
VectorDrawable is doing is extremely helpful and necessary for understanding some of the advanced features we’ll get to later.
Paths by themselves don’t draw anything, they need to be stroked and/or filled.
Part 2 of this series goes into more detail on the different ways of filling/stroking paths.
You can also define groups of paths. This allows you to define transformations that will be applied to all paths within the group.
Note that you can’t rotate/scale/translate individual paths. If you want this behavior you’ll need to place them in a group. These transformation make little sense for static images which could ‘bake’ them into their paths directly — but they are extremely useful for animating.
You can also define
clip-paths, that is mask the area that other paths in the same group can draw to. They’re defined exactly the same way as
One limitation of note is that clip-paths are not anti-aliased.
This example (which I’ve had to enlarge greatly to show the effect) shows two approaches for drawing a camera shutter icon. The first draws the paths, the second draws a solid square, masked to the shutter shape. Masking can help to create interesting effects (especially when animated) but it’s relatively expensive so if you can avoid it by drawing a shape in a different way, then do.
Paths can be trimmed; that is only draw a subset of the entire path. You can trim filled paths but the results can be surprising! It’s more common to trim stroked paths.
You can trim either from the start, or end of a path or apply an offset to any trims. They are defined as a fraction of the path [0,1]. See how setting different trim values changes the portion of the line that is drawn. Also note that offsets can make the trim values ‘wrap around’. Once again, this property doesn’t make much sense for static images but is handy for animation.
vector element supports an
alpha property [0, 1]. Groups do not have an alpha property but individual paths support
So hopefully this post gives you an idea of what vector assets are, their benefits and trade-offs. Android’s vector format is capable and has widespread support. Given the variety of devices in the market, using vector assets should be your default choice, only resorting to rasters in special cases. Join us in the next posts to learn more:
In the previous article, we looked at Android’s VectorDrawable format, going into its benefits and capabilities.medium.com
In previous posts we’ve looked at Android’s VectorDrawable image format and what it can do:medium.com
Coming soon: Creating vector assets for Android
Coming soon: Profiling Android