DetectWithStreamAsync takes long time

Question

I find a sample code that uses Azure Face API ->https://github.com/jernejk/RealTimeFaceApi

I found DetectFace time + Recognize Facetime is too long so I tried to optimize by cropping the image to only the face (with 10-30% padding) to have <80kB upload. Although the time has become shorter, it is not "real-time". Please refer to the below Log:

This is the code segment↓

     private static async Task StartRecognizing(Mat image)  
            {  
                try  
                {  
                    // TODO: Optimize by cropping the image to only the face (with 10-30% padding) to have <50kB upload.  
                    var stream = image.ToMemoryStream();  
      
                    Console.WriteLine(DateTime.Now + ": Sending " + (stream.Length / 1024) + "kB to recognize face.");  
      
                    Stopwatch stopwatch = Stopwatch.StartNew();  
                    detectedFaces = await _faceClient.Face.DetectWithStreamAsync(stream, true, false);  
                    var faceIds = detectedFaces.Where(f => f.FaceId.HasValue).Select(f => f.FaceId.Value).ToList();  
                    
      
                    Console.WriteLine(DateTime.Now + ": Found " + faceIds.Count + " faces in " + stopwatch.ElapsedMilliseconds + "ms.");  
      
                    stopwatch.Stop();  
      
                    if (faceIds.Any())  
                    {  
                        stopwatch.Restart();  
      
                        var potentialUsers = await _faceClient.Face.IdentifyAsync(faceIds, FaceGroupId);  
      
                        stopwatch.Stop();  
      
                        Console.WriteLine(DateTime.Now + ": Recognized " + potentialUsers.Count + " candidates in " + stopwatch.ElapsedMilliseconds + "ms.");  
                        foreach (var candidate in potentialUsers.Select(u => u.Candidates.FirstOrDefault()))  
                        {  
                            var candidateName = await GetCandidateName(candidate?.PersonId);  
                            Console.WriteLine($"{DateTime.Now}: {candidateName} ({candidate?.PersonId})");  
                        }  
                    }  
                    else  
                    {  
                        Console.WriteLine(DateTime.Now + $": No clear shot on the faces.");  
                    }  
                }  
                catch (APIErrorException apiError)  
                {  
                    Console.WriteLine(DateTime.Now + ": Cognitive service error: " + apiError?.Body?.Error?.Message);  
                }  
                catch (Exception e)  
                {  
                    Console.WriteLine(DateTime.Now + ": Getting identity failed: " + e.ToString());  
                }  
      
                _faceRecognitionTask = null;  
            }  
  

 

What I want to know is
Does the speed of FaceAPI completely depend on network speed?

Answer 1

Hi, thanks for reaching out. Some possible causes for latency when using Face service include:

• Large upload size
• Slow connection between the Cognitive Service and a remote URL
• Slow connection between your compute resource and the Face service

Please refer to the following document for more details as well as ways to mitigate latency. Hope this helps.