I wrote a test application in Qt4 which uses QImage.scaled() or QPixmap.scaled() methods that turned to be very slow. Even a perspective transform is faster, while a scaling transform is the same slow.
[I tried to scale directly a QPainter but I do not master paintEvent() so I always get "painter not active" or paintEvent() is not called at all. So I do not know the painter scaling performaces.]
I ask here if the same issue is known for Quartz/Cocoa or instead their performances for similar tasks are better. I am particularly interested in native Quartz pdf rendering capability and subsequent image scaling.
NIRTimer *timer = [NIRTimer timer];
[timer start];
NSImage *image = [[NSImage alloc]initWithContentsOfFile:#"filename"];
NSImage *scaledImage = [[NSImage alloc]initWithSize:NSMakeSize(720, 480)];
[scaledImage lockFocus];
[image drawInRect:NSMakeRect(0, 0, 720, 480) fromRect:NSZeroRect operation:NSCompositeSourceAtop fraction:1];
[scaledImage unlockFocus];
[image release];
[scaledImage release];
NSLog(#"time: %ld", [timer microseconds]);
This is how to scale an image in Cocoa, and it takes 80000 microseconds (0.08 seconds).
Related
I have an NSImage, initialized with PDF data, created like this:
NSData* data = [view dataWithPDFInsideRect:view.bounds];
slideImage = [[NSImage alloc] initWithData:data];
The slideImage is now the size of the view.
When I try to render the image in an NSImageView, it only draws sharp when the image view is exactly the original size of the image, even if you clear the cache or change the image size. I tried setting the cacheMode to NSImageCacheNever, which also didn't work. The only image rep in the image is the PDF one, and when I render it to a PDF file it shows that it's vector.
As a workaround, I create a NSBitmapImageRep with a different size, call drawInRect on the original image, and put the bitmap representation inside a new NSImage and render that, which works, but it feels like it's not optimal:
- (NSBitmapImageRep*)drawToBitmapOfWidth:(NSInteger)width
andHeight:(NSInteger)height
withScale:(CGFloat)scale
{
NSBitmapImageRep *bmpImageRep = [[NSBitmapImageRep alloc]
initWithBitmapDataPlanes:NULL
pixelsWide:width * scale
pixelsHigh:height * scale
bitsPerSample:8
samplesPerPixel:4
hasAlpha:YES
isPlanar:NO
colorSpaceName:NSCalibratedRGBColorSpace
bitmapFormat:NSAlphaFirstBitmapFormat
bytesPerRow:0
bitsPerPixel:0
];
bmpImageRep = [bmpImageRep bitmapImageRepByRetaggingWithColorSpace:
[NSColorSpace sRGBColorSpace]];
[bmpImageRep setSize:NSMakeSize(width, height)];
NSGraphicsContext *bitmapContext = [NSGraphicsContext graphicsContextWithBitmapImageRep:bmpImageRep];
[NSGraphicsContext saveGraphicsState];
[NSGraphicsContext setCurrentContext:bitmapContext];
[self drawInRect:NSMakeRect(0, 0, width, height) fromRect:NSZeroRect operation:NSCompositeCopy fraction:1];
[NSGraphicsContext restoreGraphicsState];
return bmpImageRep;
}
- (NSImage*)rasterizedImageForSize:(NSSize)size
{
NSImage* newImage = [[NSImage alloc] initWithSize:size];
NSBitmapImageRep* rep = [self drawToBitmapOfWidth:size.width andHeight:size.height withScale:1];
[newImage addRepresentation:rep];
return newImage;
}
How can I get the PDF to render nicely at any size without resorting to hacks like mine?
The point of NSImage is that you create it with the size (in points) that you want it to be. The backing representation can be vector based (e.g. PDF), and the NSImage is resolution independent (i.e. it supports different pixels per point), but the NSImage still has a fixed size (in points).
One one the points of an NSImage is that it will / can add a cache representation to speed up subsequent drawing.
If you need to draw a PDF to multiple sizes, and you want to use an NSImage, you're probably best of creating an NSImage for your given target size. If you want to, you can keep the NSPDFImageRef around -- I don't think it'll save you much.
We tried the following:
NSPDFImageRep* rep = self.representations.lastObject;
return [NSImage imageWithSize:size flipped:NO drawingHandler:^BOOL (NSRect dstRect)
{
[[NSGraphicsContext currentContext] setImageInterpolation:NSImageInterpolationHigh];
[rep drawInRect:dstRect fromRect:NSZeroRect operation:NSCompositeCopy fraction:1 respectFlipped:YES hints:#{
NSImageHintInterpolation: #(NSImageInterpolationHigh)
}];
return YES;
}];
And that does give you nice results when scaling up, but makes for blurry images
when scaling down.
I am working on a OSX/Cocoa graphics application which (for performance reasons) I would like to render at 640x480 when the user selects "full screen" mode. For what it's worth, the content is a custom NSView which draws using openGL.
I understand that rather than actually change the user's resolution, it's preferable to change the backbuffer (as explained on another SO question here: Programmatically change resolution OS X).
Following that advice, I end up with the following two methods (see below) to toggle between fullscreen and windowed. The trouble is that when I go fullscreen, the content does indeed render at 640x480 but is not scaled (IE it appears as if we stayed at the window's resolution and "zoomed" into a 640x480 corner of the render).
I'm probably missing something obvious here - I suppose I could translate the render according to the actual screen resolution to "center" it, but that seems overcomplicated?
- (void)goFullscreen{
// Bounce if we're already fullscreen
if(_isFullscreen){return;}
// Save original size and position
NSRect frame = [self.window.contentView frame];
original_size = frame.size;
original_position = frame.origin;
NSDictionary *options = [NSDictionary dictionaryWithObjectsAndKeys:
[NSNumber numberWithBool:NO],NSFullScreenModeAllScreens,
nil];
// In lieu of changing resolution, we set the backbuffer to 640x480
GLint dim[2] = {640, 480};
CGLSetParameter([[self openGLContext] CGLContextObj], kCGLCPSurfaceBackingSize, dim);
CGLEnable ([[self openGLContext] CGLContextObj], kCGLCESurfaceBackingSize);
// Go fullscreen!
[self enterFullScreenMode:[NSScreen mainScreen] withOptions:options];
_isFullscreen=true;
}
- (void)goWindowed{
// Bounce if we're already windowed
if(!_isFullscreen){return;}
// Reset backbuffer
GLint dim[2] = {original_size.width, original_size.height};
CGLSetParameter([[self openGLContext] CGLContextObj], kCGLCPSurfaceBackingSize, dim);
CGLEnable ([[self openGLContext] CGLContextObj], kCGLCESurfaceBackingSize);
// Go windowed!
[self exitFullScreenModeWithOptions:nil];
[self.window makeFirstResponder:self];
_isFullscreen=false;
}
Update
Here's now to do something similar to datenwolf's suggestion below, but not using openGL (useful for non-gl content).
// Render into a specific size
renderDimensions = NSMakeSize(640, 480);
NSImage *drawIntoImage = [[NSImage alloc] initWithSize:renderDimensions];
[drawIntoImage lockFocus];
[self drawViewOfSize:renderDimensions];
[drawIntoImage unlockFocus];
[self syphonSendImage:drawIntoImage];
// Resize to fit preview area and draw
NSSize newSize = NSMakeSize(self.frame.size.width, self.frame.size.height);
[drawIntoImage setSize: newSize];
[[NSColor blackColor] set];
[self lockFocus];
[NSBezierPath fillRect:self.frame];
[drawIntoImage drawAtPoint:NSZeroPoint fromRect:self.frame operation:NSCompositeCopy fraction:1];
[self unlockFocus];
Use a FBO with a texture of the desired target resolution attached and render to that FBO/texture in said resolution. Then switch to the main framebuffer and draw a full screen quad using the texture rendered to just before. Use whatever magnification filter you like best. If you want to bring out the big guns you could implement a Lancosz / sinc interpolator in the fragment shader to upscaling the intermediary texture.
I am creating small NSImageViews (32x32 pixels) from large images often 512x512 or even 4096 x 2048 for an extreme test case. My problem is that with my extreme test case, my applicaiton memory footprint seems to go up by over 15MB when I display my thumbnail, this makes me think the NSImage is being stored in memory as a 4096x2048 instead of 32x32 and I was wondering if there is a way to avoid this. Here is the process I go through to create the NsImageView:
• First I create an NSImage using initByReferencingFile: (pointing to the 4096x2048 .png file)
• Next I initialize the NSImageView with a call to initWithFrame:
• Then I call setImage: to assign my NSImage to the NSImageView
• Finally I set the NSImageView to NSScaleProportionally
I clearly do nothing to force the NSImage to size down to 32x32 but I have had trouble finding a good way to handle this.
You can simply create a new 32x32 NSImage from the original and then release the original image.
First, create the 32x32 image:
NSImage *smallImage = [[NSImage alloc]initWithSize:NSMakeSize(32, 32)];
Then, lock focus on the image and draw the original on to it:
NSSize originalSize = [originalImage size];
NSRect fromRect = NSMakeRect(0, 0, originalSize.width, originalSize.height);
[smallImage lockFocus];
[originalImage drawInRect:NSMakeRect(0, 0, 32, 32) fromRect:fromRect operation:NSCompositeCopy fraction:1.0f];
[smallImage unlockFocus];
Then you may do as you please with the smaller image:
[imageView setImage:smallImage];
Remember to release!
[originalImage release];
[smallImage release];
I have an NSBitmapImageRep that I am creating the following way:
+ (NSBitmapImageRep *)bitmapRepOfImage:(NSURL *)imageURL {
CIImage *anImage = [CIImage imageWithContentsOfURL:imageURL];
CGRect outputExtent = [anImage extent];
NSBitmapImageRep *theBitMapToBeSaved = [[NSBitmapImageRep alloc]
initWithBitmapDataPlanes:NULL pixelsWide:outputExtent.size.width
pixelsHigh:outputExtent.size.height bitsPerSample:8 samplesPerPixel:4
hasAlpha:YES isPlanar:NO colorSpaceName:NSDeviceRGBColorSpace
bytesPerRow:0 bitsPerPixel:0];
NSGraphicsContext *nsContext = [NSGraphicsContext graphicsContextWithBitmapImageRep:theBitMapToBeSaved];
[NSGraphicsContext saveGraphicsState];
[NSGraphicsContext setCurrentContext: nsContext];
CGPoint p = CGPointMake(0.0, 0.0);
[[nsContext CIContext] drawImage:anImage atPoint:p fromRect:outputExtent];
[NSGraphicsContext restoreGraphicsState];
return [[theBitMapToBeSaved retain] autorelease];
}
And being saved as BMP this way:
NSBitmapImageRep *original = [imageTools bitmapRepOfImage:fileURL];
NSData *converted = [original representationUsingType:NSBMPFileType properties:nil];
[converted writeToFile:filePath atomically:YES];
The thing here is that the BMP file can be read and manipulated correctly under Mac OSX, but under Windows, it just fails to load, just like in this screenshot:
screenshot http://dl.dropbox.com/u/1661304/Grab/74a6dadb770654213cdd9290f0131880.png
If the file is opened with MS Paint (yes, MS Paint can open it) and then resaved, though, it will work.
Would appreciate a hand here. :)
Thanks in advance.
I think the main reason your code is failing is that you are creating your NSBitmapImageRep with 0 bits per pixel. That means your image rep will have precisely zero information in it. You almost certainly want 32 bits per pixel.
However, your code is an unbelievably convoluted way to obtain an NSBitmapImageRep from an image file on disk. Why on earth are you using a CIImage? That is a Core Image object designed for use with Core Image filters and makes no sense here at all. You should be using an NSImage or CGImageRef.
Your method is also poorly named. It should instead be named something like +bitmapRepForImageFileAtURL: to better indicate what it is doing.
Also, this code makes no sense:
[[theBitMapToBeSaved retain] autorelease]
Calling retain and then autorelease does nothing, because all it does in increment the retain count and then decrement it again immediately.
You are responsible for releasing theBitMapToBeSaved because you created it using alloc. Since it is being returned, you should call autorelease on it. Your additional retain call just causes a leak for no reason.
Try this:
+ (NSBitmapImageRep*)bitmapRepForImageFileAtURL:(NSURL*)imageURL
{
NSImage* image = [[[NSImage alloc] initWithContentsOfURL:imageURL] autorelease];
return [NSBitmapImageRep imageRepWithData:[image TIFFRepresentation]];
}
+ (NSData*)BMPDataForImageFileAtURL:(NSURL*)imageURL
{
NSBitmapImageRep* bitmap = [self bitmapRepForImageFileAtURL:imageURL];
return [bitmap representationUsingType:NSBMPFileType properties:nil];
}
You really need to review the Cocoa Drawing Guide and the Memory Management Guidelines, because it appears that you are having trouble with some basic concepts.
This is my image resize code:
CALayer *newCALayer = [[CALayer layer] retain];
NSImage* image = [[NSImage alloc] initWithData:[NSData dataWithContentsOfFile:path]];
CGImageRef newCGImageFullResolution = [image CGImageForProposedRect:nil context:nil hints:nil];
CGContextRef context = CGBitmapContextCreate(NULL, drawRect.size.width, drawRect.size.height,
CGImageGetBitsPerComponent(newCGImageFullResolution),
CGImageGetBytesPerRow(newCGImageFullResolution),
CGImageGetColorSpace(newCGImageFullResolution),
CGImageGetAlphaInfo(newCGImageFullResolution));
CGContextDrawImage(context, CGRectMake(0, 0, drawRect.size.width, drawRect.size.height), newCGImageFullResolution);
CGImageRef scaledImage = CGBitmapContextCreateImage(context);
newCALayer.contents = (id)scaledImage;
CGImageRelease(scaledImage);
newCALayer.contentsGravity = kCAGravityResizeAspect;
newCALayer.opacity = 0.0;
newCALayer.anchorPoint = CGPointMake(0.0f,0.0f);
newCALayer.frame = CGRectMake( 0.0,
0.0,
[Singleton sharedSingleton].fullscreenRect.size.width,
[Singleton sharedSingleton].fullscreenRect.size.height);
[newCALayer setAutoresizingMask:kCALayerWidthSizable | kCALayerHeightSizable];
//CGImageRelease(cgImageFullResolution); (bonus points if you can explain why I can't release this! I mean, I can release the scaled image ok??)
CGContextRelease(context);
[image release];
I am doing all of this from a background thread in order to preload pictures so my GUI feels snappy. It took some work getting synchronization and what not set up so the CALayers ends up in view.
But I believe the term for describing how fast this is would be "it's a dog".
Comparing to IKImageView - that thing flings up thumbnails of images faster than I can scroll.
Does anybody have some suggestions for how to handle this better than I am doing it now?
In other words, my problem is that I want to have a super-fast UX. I believe the way to accomplish this is by preloading things to CALayers (this may be wrong? I tried NSImageView and some IK-stuff, but at least CALayer is better than that).
ImageKit is probably using CGImageSourceCreateThumbnailAtIndex() to quickly get an image appropriate to the destination, rather than reading in the entire image file.
Here:
NSImage *image = [[[NSImage alloc] initWithContentsOfFile:path] autorelease];
[image setScalesWhenResized:YES]; // *
[image setDataRetained:YES]; // *
[image setSize:desiredNewSize];
Then use the image as it is.
As for why your app is slow, run it under Instruments. That will tell you specifically where you are spending the majority of the processor time you use—it may not be in your scaling code after all.
*Since 10.6, these messages do nothing useful and are deprecated, so you can omit them if you are requiring Snow Leopard or later.