Add support to PaddleOCR V6 models, add image orentation model

Author: sdcb

build/00-common.linq

@@ -23,8 +23,8 @@ static ProjectVersion[] Projects = new[]
 {
 	new ProjectVersion("Sdcb.OpenVINO", "0.7.1"),
 	new ProjectVersion("Sdcb.OpenVINO.Extensions.OpenCvSharp4", "0.7.0"),
-	new ProjectVersion("Sdcb.OpenVINO.PaddleOCR", "0.7.0"),
-	new ProjectVersion("Sdcb.OpenVINO.PaddleOCR.Models.Online", "0.7.1"),
+	new ProjectVersion("Sdcb.OpenVINO.PaddleOCR", "0.8.0"),
+	new ProjectVersion("Sdcb.OpenVINO.PaddleOCR.Models.Online", "0.8.0"),
 };
 
 static async Task DownloadFile(Uri uri, string localFile, CancellationToken cancellationToken = default)

projects/PaddleOCR/Sdcb.OpenVINO.PaddleOCR.Models.Online/Details/Utils.cs

@@ -52,12 +52,15 @@ public static async Task DownloadFiles(Uri[] uris, string localFile, Cancellatio
         throw new Exception($"Failed to download {localFile} from all uris: {string.Join(", ", uris.Select(x => x.ToString()))}");
     }
 
-    public static async Task DownloadAndExtractAsync(string name, Uri uri, string rootDir, CancellationToken cancellationToken)
+    public static async Task DownloadAndExtractAsync(string name, Uri uri, string rootDir, CancellationToken cancellationToken, params string[] expectedFileNames)
     {
         Directory.CreateDirectory(rootDir);
-        string paramsFile = Path.Combine(rootDir, "inference.pdiparams");
+        if (expectedFileNames == null || expectedFileNames.Length == 0)
+        {
+            expectedFileNames = new[] { "inference.pdiparams", "inference.pdmodel" };
+        }
 
-        if (!File.Exists(paramsFile))
+        if (!CheckLocalModel(rootDir, expectedFileNames, throwOnError: false))
         {
             string localTarFile = Path.Combine(rootDir, uri.Segments.Last());
             if (!File.Exists(localTarFile) || new FileInfo(localTarFile).Length == 0)
@@ -94,7 +97,7 @@ public static async Task DownloadAndExtractAsync(string name, Uri uri, string ro
                     archive.WriteToDirectory(rootDir);
                 }
 
-                CheckLocalOCRModel(rootDir);
+                CheckLocalModel(rootDir, expectedFileNames, throwOnError: true);
             }
 
             File.Delete(localTarFile);
@@ -193,26 +196,29 @@ private static string GetDestinationPath(string rootDir, string relativePath)
 
     public static void CheckLocalOCRModel(string rootDir)
     {
-        string[] filesToCheck = new[]
-        {
-            Path.Combine(rootDir, "inference.pdiparams"),
-            Path.Combine(rootDir, "inference.pdmodel"), 
-        };
+        CheckLocalModel(rootDir, new[] { "inference.pdiparams", "inference.pdmodel" }, throwOnError: true);
+    }
 
-        foreach (string path in filesToCheck)
+    private static bool CheckLocalModel(string rootDir, string[] expectedFileNames, bool throwOnError)
+    {
+        foreach (string fileName in expectedFileNames)
         {
-            string fileName = Path.GetFileName(path);
+            string path = Path.Combine(rootDir, fileName);
 
             if (!File.Exists(path))
             {
+                if (!throwOnError) return false;
                 throw new Exception($"{fileName} not found in {rootDir}, model error?");
             }
 
             if (new FileInfo(path).Length == 0)
             {
+                if (!throwOnError) return false;
                 throw new Exception($"{fileName} invalid(length = 0), model error?");
             }
         }
+
+        return true;
     }
 
     public readonly static Type RootType = typeof(Settings);

projects/PaddleOCR/Sdcb.OpenVINO.PaddleOCR.Models.Online/IOnlineModels.cs

@@ -0,0 +1,49 @@
+using Sdcb.OpenVINO.PaddleOCR.Models;
+using System.Threading;
+using System.Threading.Tasks;
+
+namespace Sdcb.OpenVINO.PaddleOCR.Models.Online;
+
+/// <summary>
+/// Represents an online detection model.
+/// </summary>
+public interface IOnlineDetectionModel
+{
+    /// <summary>
+    /// Downloads the model asynchronously.
+    /// </summary>
+    Task<DetectionModel> DownloadAsync(CancellationToken cancellationToken = default);
+}
+
+/// <summary>
+/// Represents an online text line orientation classification model.
+/// </summary>
+public interface IOnlineClassificationModel
+{
+    /// <summary>
+    /// Downloads the model asynchronously.
+    /// </summary>
+    Task<ClassificationModel> DownloadAsync(CancellationToken cancellationToken = default);
+}
+
+/// <summary>
+/// Represents an online recognition model.
+/// </summary>
+public interface IOnlineRecognizationModel
+{
+    /// <summary>
+    /// Downloads the model asynchronously.
+    /// </summary>
+    Task<RecognizationModel> DownloadAsync(CancellationToken cancellationToken = default);
+}
+
+/// <summary>
+/// Represents an online document orientation classification model.
+/// </summary>
+public interface IOnlineDocumentOrientationClassificationModel
+{
+    /// <summary>
+    /// Downloads the model asynchronously.
+    /// </summary>
+    Task<DocumentOrientationClassificationModel> DownloadAsync(CancellationToken cancellationToken = default);
+}

projects/PaddleOCR/Sdcb.OpenVINO.PaddleOCR.Models.Online/LocalDictOnlineRecognizationModel.cs

@@ -13,7 +13,7 @@ namespace Sdcb.OpenVINO.PaddleOCR.Models.Online;
 /// <remarks>
 /// Used for downloading and extracting a model from a url, and creating a new StreamDictFileRecognizationModel with the downloaded contents.
 /// </remarks>
-public record LocalDictOnlineRecognizationModel(string Name, string DictName, Uri Uri, ModelVersion Version)
+public record LocalDictOnlineRecognizationModel(string Name, string DictName, Uri Uri, ModelVersion Version) : IOnlineRecognizationModel
 {
     /// <summary>
     /// Gets or sets the root directory for the downloaded models.

projects/PaddleOCR/Sdcb.OpenVINO.PaddleOCR.Models.Online/OnlineClassificationModel.cs

@@ -10,7 +10,7 @@ namespace Sdcb.OpenVINO.PaddleOCR.Models.Online;
 /// <summary>
 /// Represents an online classification model that can be downloaded and used for text angle classification.
 /// </summary>
-public record OnlineClassificationModel(string Name, Uri Uri, ModelVersion Version)
+public record OnlineClassificationModel(string Name, Uri Uri, ModelVersion Version) : IOnlineClassificationModel
 {
     /// <summary>
     /// Gets the root directory of the model.
@@ -28,6 +28,11 @@ public async Task<FileClassificationModel> DownloadAsync(CancellationToken cance
         return new FileClassificationModel(RootDirectory, Version);
     }
 
+    async Task<ClassificationModel> IOnlineClassificationModel.DownloadAsync(CancellationToken cancellationToken)
+    {
+        return await DownloadAsync(cancellationToken);
+    }
+
     /// <summary>
     /// Gets an online classification model for slim quantized text angle classification (Size: 2.1M).
     /// </summary>

projects/PaddleOCR/Sdcb.OpenVINO.PaddleOCR.Models.Online/OnlineDetectionModel.cs

@@ -10,7 +10,7 @@ namespace Sdcb.OpenVINO.PaddleOCR.Models.Online;
 /// <summary>
 /// Represents a model for online object detection.
 /// </summary>
-public record OnlineDetectionModel(string Name, Uri Uri, ModelVersion Version)
+public record OnlineDetectionModel(string Name, Uri Uri, ModelVersion Version) : IOnlineDetectionModel
 {
     /// <summary>
     /// Gets the root directory of the model.
@@ -29,6 +29,11 @@ public async Task<FileDetectionModel> DownloadAsync(CancellationToken cancellati
         return new FileDetectionModel(RootDirectory, Version);
     }
 
+    async Task<DetectionModel> IOnlineDetectionModel.DownloadAsync(CancellationToken cancellationToken)
+    {
+        return await DownloadAsync(cancellationToken);
+    }
+
     /// <summary>
     /// [New] v4 model, supporting Chinese, English, multilingual text detection
     /// (Size: 4.66M)

projects/PaddleOCR/Sdcb.OpenVINO.PaddleOCR.Models.Online/OnlineDocumentOrientationClassificationModel.cs

@@ -0,0 +1,39 @@
+using Sdcb.OpenVINO.PaddleOCR.Models;
+using Sdcb.OpenVINO.PaddleOCR.Models.Details;
+using Sdcb.OpenVINO.PaddleOCR.Models.Online.Details;
+using System;
+using System.IO;
+using System.Threading;
+using System.Threading.Tasks;
+
+namespace Sdcb.OpenVINO.PaddleOCR.Models.Online;
+
+/// <summary>
+/// Represents an online ONNX document orientation classification model.
+/// </summary>
+public record OnlineDocumentOrientationClassificationModel(string Name, Uri Uri) : IOnlineDocumentOrientationClassificationModel
+{
+    /// <summary>
+    /// Gets the root directory of the model.
+    /// </summary>
+    public string RootDirectory => Path.Combine(Settings.GlobalModelDirectory, Name);
+
+    /// <summary>
+    /// Downloads and extracts the model asynchronously.
+    /// </summary>
+    public async Task<FileDocumentOrientationClassificationModel> DownloadAsync(CancellationToken cancellationToken = default)
+    {
+        await Utils.DownloadAndExtractAsync(Name, Uri, RootDirectory, cancellationToken, "inference.onnx", "inference.yml");
+        return new FileDocumentOrientationClassificationModel(RootDirectory);
+    }
+
+    async Task<DocumentOrientationClassificationModel> IOnlineDocumentOrientationClassificationModel.DownloadAsync(CancellationToken cancellationToken)
+    {
+        return await DownloadAsync(cancellationToken);
+    }
+
+    /// <summary>
+    /// PP-LCNet x1.0 document orientation model.
+    /// </summary>
+    public static OnlineDocumentOrientationClassificationModel PPDocOrientationX10 => new("PP-LCNet_x1_0_doc_ori_onnx_infer", new Uri("https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_onnx_infer.tar"));
+}

projects/PaddleOCR/Sdcb.OpenVINO.PaddleOCR.Models.Online/OnlineFullModels.cs

@@ -10,7 +10,11 @@ namespace Sdcb.OpenVINO.PaddleOCR.Models.Online;
 /// <remarks>
 /// Contains static instances for commonly used models and a method for downloading a full OCR model.
 /// </remarks>
-public record class OnlineFullModels(OnlineDetectionModel DetModel, OnlineClassificationModel? ClsModel, LocalDictOnlineRecognizationModel RecModel)
+public record class OnlineFullModels(
+    IOnlineDetectionModel DetModel,
+    IOnlineClassificationModel? ClsModel,
+    IOnlineRecognizationModel RecModel,
+    IOnlineDocumentOrientationClassificationModel? DocOrientationModel = null)
 {
     /// <summary>
     /// Downloads a full OCR model asynchronously.
@@ -19,12 +23,40 @@ public record class OnlineFullModels(OnlineDetectionModel DetModel, OnlineClassi
     /// <returns><see cref="FullOcrModel"/> instance.</returns>
     public async Task<FullOcrModel> DownloadAsync(CancellationToken cancellationToken = default)
     {
-        FileDetectionModel localDetModel = await DetModel.DownloadAsync(cancellationToken);
-        FileClassificationModel? localClsModel = ClsModel != null ? await ClsModel.DownloadAsync(cancellationToken) : null;
+        DetectionModel localDetModel = await DetModel.DownloadAsync(cancellationToken);
+        ClassificationModel? localClsModel = ClsModel != null ? await ClsModel.DownloadAsync(cancellationToken) : null;
         RecognizationModel localRecModel = await RecModel.DownloadAsync(cancellationToken);
-        return new FullOcrModel(localDetModel, localClsModel, localRecModel);
+        DocumentOrientationClassificationModel? localDocOrientationModel = DocOrientationModel != null ? await DocOrientationModel.DownloadAsync(cancellationToken) : null;
+        return new FullOcrModel(localDetModel, localClsModel, localRecModel, localDocOrientationModel);
     }
 
+    /// <summary>
+    /// The Chinese PP-OCRv6 medium ONNX version.
+    /// </summary>
+    public readonly static OnlineFullModels ChineseV6Medium = new(
+        OnlineOnnxDetectionModel.ChineseV6Medium,
+        OnlineOnnxClassificationModel.TextLineOrientationX10,
+        OnlineOnnxRecognizationModel.ChineseV6Medium,
+        OnlineDocumentOrientationClassificationModel.PPDocOrientationX10);
+
+    /// <summary>
+    /// The Chinese PP-OCRv6 small ONNX version.
+    /// </summary>
+    public readonly static OnlineFullModels ChineseV6Small = new(
+        OnlineOnnxDetectionModel.ChineseV6Small,
+        OnlineOnnxClassificationModel.TextLineOrientationX025,
+        OnlineOnnxRecognizationModel.ChineseV6Small,
+        OnlineDocumentOrientationClassificationModel.PPDocOrientationX10);
+
+    /// <summary>
+    /// The Chinese PP-OCRv6 tiny ONNX version.
+    /// </summary>
+    public readonly static OnlineFullModels ChineseV6Tiny = new(
+        OnlineOnnxDetectionModel.ChineseV6Tiny,
+        OnlineOnnxClassificationModel.TextLineOrientationX025,
+        OnlineOnnxRecognizationModel.ChineseV6Tiny,
+        OnlineDocumentOrientationClassificationModel.PPDocOrientationX10);
+
     /// <summary>
     /// The Chinese V4 version.
     /// </summary>

projects/PaddleOCR/Sdcb.OpenVINO.PaddleOCR.Models.Online/OnlineOnnxClassificationModel.cs

@@ -0,0 +1,44 @@
+using Sdcb.OpenVINO.PaddleOCR.Models;
+using Sdcb.OpenVINO.PaddleOCR.Models.Details;
+using Sdcb.OpenVINO.PaddleOCR.Models.Online.Details;
+using System;
+using System.IO;
+using System.Threading;
+using System.Threading.Tasks;
+
+namespace Sdcb.OpenVINO.PaddleOCR.Models.Online;
+
+/// <summary>
+/// Represents an online ONNX text line orientation classification model.
+/// </summary>
+public record OnlineOnnxClassificationModel(string Name, Uri Uri, ModelVersion Version, NCHW Shape) : IOnlineClassificationModel
+{
+    /// <summary>
+    /// Gets the root directory of the model.
+    /// </summary>
+    public string RootDirectory => Path.Combine(Settings.GlobalModelDirectory, Name);
+
+    /// <summary>
+    /// Downloads and extracts the model asynchronously.
+    /// </summary>
+    public async Task<FileOnnxClassificationModel> DownloadAsync(CancellationToken cancellationToken = default)
+    {
+        await Utils.DownloadAndExtractAsync(Name, Uri, RootDirectory, cancellationToken, "inference.onnx", "inference.yml");
+        return new FileOnnxClassificationModel(RootDirectory, Version, Shape);
+    }
+
+    async Task<ClassificationModel> IOnlineClassificationModel.DownloadAsync(CancellationToken cancellationToken)
+    {
+        return await DownloadAsync(cancellationToken);
+    }
+
+    /// <summary>
+    /// PP-LCNet x0.25 text line orientation model.
+    /// </summary>
+    public static OnlineOnnxClassificationModel TextLineOrientationX025 => new("PP-LCNet_x0_25_textline_ori_onnx_infer", new Uri("https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x0_25_textline_ori_onnx_infer.tar"), ModelVersion.V6, new NCHW(-1, 3, 80, 160));
+
+    /// <summary>
+    /// PP-LCNet x1.0 text line orientation model.
+    /// </summary>
+    public static OnlineOnnxClassificationModel TextLineOrientationX10 => new("PP-LCNet_x1_0_textline_ori_onnx_infer", new Uri("https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_textline_ori_onnx_infer.tar"), ModelVersion.V6, new NCHW(-1, 3, 80, 160));
+}

projects/PaddleOCR/Sdcb.OpenVINO.PaddleOCR.Models.Online/OnlineOnnxDetectionModel.cs

@@ -0,0 +1,49 @@
+using Sdcb.OpenVINO.PaddleOCR.Models;
+using Sdcb.OpenVINO.PaddleOCR.Models.Details;
+using Sdcb.OpenVINO.PaddleOCR.Models.Online.Details;
+using System;
+using System.IO;
+using System.Threading;
+using System.Threading.Tasks;
+
+namespace Sdcb.OpenVINO.PaddleOCR.Models.Online;
+
+/// <summary>
+/// Represents an online ONNX detection model.
+/// </summary>
+public record OnlineOnnxDetectionModel(string Name, Uri Uri, ModelVersion Version) : IOnlineDetectionModel
+{
+    /// <summary>
+    /// Gets the root directory of the model.
+    /// </summary>
+    public string RootDirectory => Path.Combine(Settings.GlobalModelDirectory, Name);
+
+    /// <summary>
+    /// Downloads and extracts the model files to the root directory asynchronously.
+    /// </summary>
+    public async Task<FileOnnxDetectionModel> DownloadAsync(CancellationToken cancellationToken = default)
+    {
+        await Utils.DownloadAndExtractAsync(Name, Uri, RootDirectory, cancellationToken, "inference.onnx");
+        return new FileOnnxDetectionModel(RootDirectory, Version);
+    }
+
+    async Task<DetectionModel> IOnlineDetectionModel.DownloadAsync(CancellationToken cancellationToken)
+    {
+        return await DownloadAsync(cancellationToken);
+    }
+
+    /// <summary>
+    /// PP-OCRv6 medium detection model.
+    /// </summary>
+    public static OnlineOnnxDetectionModel ChineseV6Medium => new("PP-OCRv6_medium_det_onnx_infer", new Uri("https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_medium_det_onnx_infer.tar"), ModelVersion.V6);
+
+    /// <summary>
+    /// PP-OCRv6 small detection model.
+    /// </summary>
+    public static OnlineOnnxDetectionModel ChineseV6Small => new("PP-OCRv6_small_det_onnx_infer", new Uri("https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_small_det_onnx_infer.tar"), ModelVersion.V6);
+
+    /// <summary>
+    /// PP-OCRv6 tiny detection model.
+    /// </summary>
+    public static OnlineOnnxDetectionModel ChineseV6Tiny => new("PP-OCRv6_tiny_det_onnx_infer", new Uri("https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_tiny_det_onnx_infer.tar"), ModelVersion.V6);
+}