maddo/Catalog
1
0
Fork 0
Code Issues Pull requests 1 Projects Releases 2 Packages Wiki Activity Actions

feat: Enhance AI extraction summaries and worker allocation for GPU support
Some checks failed
Build Windows Avalonia / build (push) Failing after 1m38s
Details
Build Windows Avalonia / release (push) Has been skipped
Details

Browse source
This commit is contained in:
MaddoScientisto 2026-05-09 19:31:21 +02:00
commit f57dc1edba
3 changed files with 90 additions and 51 deletions
Download patch file Download diff file Expand all files Collapse all files

128
imagecatalog/Services/AiExtractionService.cs
View file

@ -45,8 +45,8 @@ public class AiExtractionService : IAiExtractionService
var total = imageFiles.Count;
if (total == 0)
{
var emptySummary = new AiExtractionRunSummary(0, 0, 0, 0, workloadLevel, workerCount);
await onProgress(new AiExtractionProgressUpdate(0, 0, 100, 0, workloadLevel, workerCount)).ConfigureAwait(false);
var emptySummary = new AiExtractionRunSummary(0, 0, 0, 0, workloadLevel, workerCount, request.UseGpu);
await onProgress(new AiExtractionProgressUpdate(0, 0, 100, 0, workloadLevel, workerCount, request.UseGpu)).ConfigureAwait(false);
return emptySummary;
}
@ -79,7 +79,7 @@ public class AiExtractionService : IAiExtractionService
var currentProcessed = Interlocked.Increment(ref processed);
var averageImagesPerSecond = CalculateAverageImagesPerSecond(currentProcessed, stopwatch.Elapsed);
var percent = currentProcessed * 100.0 / total;
await onProgress(new AiExtractionProgressUpdate(total, currentProcessed, percent, averageImagesPerSecond, workloadLevel, workerCount)).ConfigureAwait(false);
await onProgress(new AiExtractionProgressUpdate(total, currentProcessed, percent, averageImagesPerSecond, workloadLevel, workerCount, request.UseGpu)).ConfigureAwait(false);
var shouldLog = false;
lock (logLock)
@ -94,54 +94,74 @@ public class AiExtractionService : IAiExtractionService
if (shouldLog)
{
_logger.LogInformation(
"Number AI progress: {Processed}/{Total} ({Percent:F1}%), {ImagesPerSecond:F2} img/s avg, workload {WorkloadLevel} ({WorkerCount} workers)",
"Number AI progress: {Processed}/{Total} ({Percent:F1}%), {ImagesPerSecond:F2} img/s avg, workload {WorkloadLevel} ({WorkerCount} {ExecutionUnit})",
currentProcessed,
total,
percent,
averageImagesPerSecond,
workloadLevel,
workerCount);
workerCount,
request.UseGpu ? "batch" : "workers");
}
}
}, token);
var workerTasks = Enumerable.Range(0, workerCount)
.Select(_ => Task.Run(async () =>
{
using var engine = new NumberRecognitionEngine(modelConfiguration, _logger);
await foreach (var file in fileChannel.Reader.ReadAllAsync(token).ConfigureAwait(false))
{
var extracted = string.Empty;
try
{
extracted = engine.ProcessImage(file).Text;
}
catch (Exception ex)
{
lock (failureLock)
{
failed++;
firstFailure ??= ex;
}
_logger.LogWarning(ex, "Error processing AI OCR for {File}", file);
}
await resultChannel.Writer.WriteAsync(new AiResultItem { Path = file, Text = extracted }, token).ConfigureAwait(false);
}
}, token))
.ToArray();
try
{
foreach (var file in imageFiles)
if (request.UseGpu)
{
await fileChannel.Writer.WriteAsync(file, token).ConfigureAwait(false);
}
using var engine = new NumberRecognitionEngine(modelConfiguration, _logger);
var resultProgress = new SynchronousProgress<ImageResult>(result =>
{
resultChannel.Writer.TryWrite(new AiResultItem { Path = result.FilePath, Text = result.Text });
});
fileChannel.Writer.TryComplete();
await Task.WhenAll(workerTasks).ConfigureAwait(false);
await engine.ProcessFilesAsync(
imageFiles,
skipTextNegative: false,
maxDegreeOfParallelism: workerCount,
progress: null,
resultProgress: resultProgress,
cancellationToken: token).ConfigureAwait(false);
}
else
{
var workerTasks = Enumerable.Range(0, workerCount)
.Select(_ => Task.Run(async () =>
{
using var engine = new NumberRecognitionEngine(modelConfiguration, _logger);
await foreach (var file in fileChannel.Reader.ReadAllAsync(token).ConfigureAwait(false))
{
var extracted = string.Empty;
try
{
extracted = engine.ProcessImage(file).Text;
}
catch (Exception ex)
{
lock (failureLock)
{
failed++;
firstFailure ??= ex;
}
_logger.LogWarning(ex, "Error processing AI OCR for {File}", file);
}
await resultChannel.Writer.WriteAsync(new AiResultItem { Path = file, Text = extracted }, token).ConfigureAwait(false);
}
}, token))
.ToArray();
foreach (var file in imageFiles)
{
await fileChannel.Writer.WriteAsync(file, token).ConfigureAwait(false);
}
fileChannel.Writer.TryComplete();
await Task.WhenAll(workerTasks).ConfigureAwait(false);
}
}
finally
{
@ -161,16 +181,18 @@ public class AiExtractionService : IAiExtractionService
failed,
CalculateAverageImagesPerSecond(processed, stopwatch.Elapsed),
workloadLevel,
workerCount);
workerCount,
request.UseGpu);
_logger.LogInformation(
"Number AI completed: {Processed}/{Total} processed, {Failed} failures, {ImagesPerSecond:F2} img/s avg, workload {WorkloadLevel} ({WorkerCount} workers)",
"Number AI completed: {Processed}/{Total} processed, {Failed} failures, {ImagesPerSecond:F2} img/s avg, workload {WorkloadLevel} ({WorkerCount} {ExecutionUnit})",
summary.ProcessedFiles,
summary.TotalFiles,
summary.FailedFiles,
summary.AverageImagesPerSecond,
summary.WorkloadLevel,
summary.WorkerCount);
summary.WorkerCount,
request.UseGpu ? "batch" : "workers");
if (!string.IsNullOrWhiteSpace(request.CsvOutputPath))
{
@ -217,11 +239,11 @@ public class AiExtractionService : IAiExtractionService
var requestedWorkers = useGpu
? normalized switch
{
1 => 2,
2 => 4,
3 => 8,
4 => 12,
_ => 16
1 => 4,
2 => 8,
3 => 16,
4 => 24,
_ => 32
}
: normalized switch
{
@ -232,7 +254,19 @@ public class AiExtractionService : IAiExtractionService
_ => 5
};
return Math.Min(requestedWorkers, maxWorkers);
return useGpu ? requestedWorkers : Math.Min(requestedWorkers, maxWorkers);
}
private sealed class SynchronousProgress<T> : IProgress<T>
{
private readonly Action<T> _handler;
public SynchronousProgress(Action<T> handler)
{
_handler = handler;
}
public void Report(T value) => _handler(value);
}
private static ModelConfiguration BuildModelConfiguration(string modelsFolderPath, bool useGpu)