4 ms·

TIFF supports many useful features for processing very large images. Tiled compression, storing channels in contiguous hyperplanes or on a single image plane,

by cjhanks 10y ago

TIFF supports many useful features for processing very large images. Tiled compression, storing channels in contiguous hyperplanes or on a single image plane, multiple levels of detail, custom compression codecs (lossless or lossy), sparse images, arbitrary bit widths, and permits storing arbitrary metadata with the data. Viewers of libtiff often treat it like PNG/JPEG, but good Tiff viewers can leverage this functionality. Typically few batteries are included. Tiff is kind of a cross breed between structured portable formats (HDF5/NetCDF) and imagery.

banachtarski 10y ago
I don't doubt that you can have a library that provides all those features. All those features can exist for any lossless image format. My point is that as far as comparing image formats goes, the capabilities of the frontend library isn't a useful metric for evaluating the format itself which, in my opinion, can optimize for the following traits: 1. On-disk size 2. Compression/decompression speed 3. Access speed (for use in image analysis algorithms, editing, compositing, etc) 4. GPU friendliness (which operates in warps on small blocks of contiguous pixels in image space) TIFF doesn't really optimize for any of these (not even #2 since in-memory decompression can be faster than paging uncompressed data from disk)
- rleigh 10y ago
  TIFF is the standard format for scientific imaging. It's the container format used by well over 80% of all imaging formats out there. It's being slowly replaced by HDF5 for some applications. But all of the features you mention above are tunable and are in practice perfectly fine with TIFF containers.
- 7952 10y ago
  The big problem with highly tuned formats is that they make a lot of choices for the user. This works well when the format is used exactly as designed, but that is often not the case. With a tif I can pick exactly the compression, bit depth, color mode, tiling etc. that I need for the specific data.
- gpvos 10y ago
  > All those features can exist for any lossless image format. Yes, but for most image formats they aren't defined. They are for TIFF. The problem, of course, is that if you use an exotic TIFF feature, very few TIFF readers will understand you.

by cjhanks 10y ago

banachtarski 10y ago

I don't doubt that you can have a library that provides all those features. All those features can exist for any lossless image format. My point is that as far as comparing image formats goes, the capabilities of the frontend library isn't a useful metric for evaluating the format itself which, in my opinion, can optimize for the following traits: 1. On-disk size 2. Compression/decompression speed 3. Access speed (for use in image analysis algorithms, editing, compositing, etc) 4. GPU friendliness (which operates in warps on small blocks of contiguous pixels in image space) TIFF doesn't really optimize for any of these (not even #2 since in-memory decompression can be faster than paging uncompressed data from disk)

rleigh 10y ago
TIFF is the standard format for scientific imaging. It's the container format used by well over 80% of all imaging formats out there. It's being slowly replaced by HDF5 for some applications. But all of the features you mention above are tunable and are in practice perfectly fine with TIFF containers.
7952 10y ago
The big problem with highly tuned formats is that they make a lot of choices for the user. This works well when the format is used exactly as designed, but that is often not the case. With a tif I can pick exactly the compression, bit depth, color mode, tiling etc. that I need for the specific data.
gpvos 10y ago
> All those features can exist for any lossless image format. Yes, but for most image formats they aren't defined. They are for TIFF. The problem, of course, is that if you use an exotic TIFF feature, very few TIFF readers will understand you.