Fast function to extract image size from JPEG file

[dgolub]

My application currently contains a custom function based on the example at https://stackoverflow.com/questions/552467/how-do-i-reliably-get-an-image-dimensions-in-net-without-loading-the-image to extract the image size from a JPEG files. It can currently get an image size in <1 ms per file. I recently migrated from System.Drawing to ImageSharp, which generally went very well. Based on the documentation, I assumed that Image.Identify would do the job and that I'd be able to eliminate the code to read JPEG files manually. However, when I made that change, the time per image went up to a little over 60 ms, which is prohibitive. It would be great if ImageSharp provided a comparably fast function for this so that users don't have to implement custom JPEG reading code to get good performance.

[JimBobSquarePants]

Something is wrong if it’s taking that long.

[antonfirsov]

Platform? Runtime? OS? Other information that may give a lead? This operation should be very fast on x86/x64 + coreclr. One possible reason can be that you are on blazor, xamarin or similar and not using AOT.

[antonfirsov]

We need an Identify overload that can disable metadata parsing. This is a tricky API design problem though, since IgnoreMetadata is a property of individual format-specific decoder types, and an Identify overload taking an IImageDecoder feels counter-intuitive. Probably good as a start though. (In the case of the current discussion for example the format is known.)

Sidenote:
@dgolub please don't take it as the lack of such overloads is your problem. As noted in #1966 (reply in thread), you should not see 60ms in a proper benchmark.

[dgolub]

Thanks for looking into this! Any idea why the test showed as running for 60 ms when I used Image.Identify but dropped to less than 1 ms when I switched it back to the code from Stack Overflow? Some sort of overhead with calling into a different assembly or something?

[antonfirsov]

Could it be related to #1809 (comment)?

[br3aker]

Any idea why the test showed as running for 60 ms

.net overhead, JIT, etc.
System.Drawing is actually native C++ GDI wrapper for .net so there's almost no overhead, that binary parsing thing has too little code for any significant overhead. ImageSharp has to compile a lot of things before the first launch but it should be a lot faster on second Identify call.

[luoxgwb]

BenchmarkDotNet=v0.13.1, OS=Windows 10.0.22000
Intel Core i7-7700 CPU 3.60GHz (Kaby Lake), 1 CPU, 8 logical and 4 physical cores
.NET SDK=6.0.200
  [Host]     : .NET 5.0.14 (5.0.1422.5710), X64 RyuJIT  [AttachedDebugger]
  DefaultJob : .NET 5.0.14 (5.0.1422.5710), X64 RyuJIT

Method	Mean	Error	StdDev	Min	Max	Gen 0	Allocated
TestIdentifyImage	207.0 μs	4.04 μs	5.11 μs	201.5 μs	217.7 μs	2.1973	79 KB