feat: add base ocr example

Author: ikesnowy

README.md

@@ -275,23 +275,23 @@ while (true)
  Console.Write("Reasoning > ");
  reasoning = true;
  }
-
+ 
  Console.Write(choice.Message.ReasoningContent);
  continue;
  }
-
+ 
  if (reasoning)
  {
  reasoning = false;
  Console.WriteLine();
  Console.Write("Assistant > ");
  }
-
+ 
  Console.Write(choice.Message.Content);
  reply.Append(choice.Message.Content);
  usage = chunk.Usage;
  }
-
+ 
  Console.WriteLine();
  messages.Add(TextChatMessage.Assistant(reply.ToString()));
  if (usage != null)
@@ -593,6 +593,63 @@ await foreach (var modelResponse in response)
 }
 ```
 
+### OCR
+
+Base example of OCR
+
+```csharp
+// upload file
+await using var tilted = File.OpenRead("tilted.png");
+var ossLink = await client.UploadTemporaryFileAsync("qwen-vl-ocr-latest", tilted, "tilted.jpg");
+Console.WriteLine($"File uploaded: {ossLink}");
+var messages = new List<MultimodalMessage>();
+messages.Add(
+ MultimodalMessage.User(
+ [
+ // set enableRotate to true if your source image is tilted.
+ MultimodalMessageContent.ImageContent(ossLink, enableRotate: true),
+ ]));
+var completion = client.GetMultimodalGenerationStreamAsync(
+ new ModelRequest<MultimodalInput, IMultimodalParameters>()
+ {
+ Model = "qwen-vl-ocr-latest",
+ Input = new MultimodalInput() { Messages = messages },
+ Parameters = new MultimodalParameters()
+ {
+ IncrementalOutput = true,
+ }
+ });
+var reply = new StringBuilder();
+var first = false;
+MultimodalTokenUsage? usage = null;
+await foreach (var chunk in completion)
+{
+ var choice = chunk.Output.Choices[0];
+ if (first)
+ {
+ first = false;
+ Console.Write("Assistant > ");
+ }
+
+ if (choice.Message.Content.Count == 0)
+ {
+ continue;
+ }
+
+ Console.Write(choice.Message.Content[0].Text);
+ reply.Append(choice.Message.Content[0].Text);
+ usage = chunk.Usage;
+}
+
+Console.WriteLine();
+messages.Add(MultimodalMessage.Assistant([MultimodalMessageContent.TextContent(reply.ToString())]));
+if (usage != null)
+{
+ Console.WriteLine(
+ $"Usage: in({usage.InputTokens})/out({usage.OutputTokens})/image({usage.ImageTokens})/total({usage.TotalTokens})");
+}
+```
+
 ## Text-to-Speech
 
 Create a speech synthesis session using `dashScopeClient.CreateSpeechSynthesizerSocketSessionAsync()`.

README.zh-Hans.md

@@ -99,6 +99,9 @@ public class YourService(IDashScopeClient client)
  - [长上下文（Qwen-Long）](#长上下文（Qwen-Long）)
 
 - [多模态](#多模态) - QWen-VL，QVQ 等，支持推理/视觉理解/OCR/音频理解等场景
+ - [视觉理解/推理](#视觉理解/推理) - 图像/视频输入与理解，支持推理模式
+ - [文字提取](#文字提取) - OCR 任务，读取表格/文档/公式等
+
 - [语音合成](#语音合成) - CosyVoice，Sambert 等，支持 TTS 等应用场景
 - [图像生成](#图像生成) - wanx2.1 等，支持文生图，人像风格重绘等应用场景
 - [应用调用](#应用调用)
@@ -2445,6 +2448,105 @@ messages.Add(
  ]));
 ```
 
+### 文字提取
+
+使用 `qwen-vl-ocr` 系列模型可以很好的完成文字提取任务，基础用法（使用本地文件）：
+
+```csharp
+// upload file
+await using var tilted = File.OpenRead("tilted.png");
+var ossLink = await client.UploadTemporaryFileAsync("qwen-vl-ocr-latest", tilted, "tilted.jpg");
+Console.WriteLine($"File uploaded: {ossLink}");
+var messages = new List<MultimodalMessage>();
+messages.Add(
+ MultimodalMessage.User(
+ [
+// 如果你的图片存在偏斜，可尝试将 enableRotate 设置为 true
+ MultimodalMessageContent.ImageContent(ossLink, enableRotate: true),
+ ]));
+var completion = client.GetMultimodalGenerationStreamAsync(
+ new ModelRequest<MultimodalInput, IMultimodalParameters>()
+ {
+ Model = "qwen-vl-ocr-latest",
+ Input = new MultimodalInput { Messages = messages },
+ Parameters = new MultimodalParameters
+ {
+ IncrementalOutput = true,
+ }
+ });
+```
+
+完整示例：
+
+```csharp
+// upload file
+await using var tilted = File.OpenRead("tilted.png");
+var ossLink = await client.UploadTemporaryFileAsync("qwen-vl-ocr-latest", tilted, "tilted.jpg");
+Console.WriteLine($"File uploaded: {ossLink}");
+var messages = new List<MultimodalMessage>();
+messages.Add(
+ MultimodalMessage.User(
+ [
+ MultimodalMessageContent.ImageContent(ossLink, enableRotate: true),
+ ]));
+var completion = client.GetMultimodalGenerationStreamAsync(
+ new ModelRequest<MultimodalInput, IMultimodalParameters>()
+ {
+ Model = "qwen-vl-ocr-latest",
+ Input = new MultimodalInput() { Messages = messages },
+ Parameters = new MultimodalParameters()
+ {
+ IncrementalOutput = true,
+ }
+ });
+var reply = new StringBuilder();
+var first = false;
+MultimodalTokenUsage? usage = null;
+await foreach (var chunk in completion)
+{
+ var choice = chunk.Output.Choices[0];
+ if (first)
+ {
+ first = false;
+ Console.Write("Assistant > ");
+ }
+
+ if (choice.Message.Content.Count == 0)
+ {
+ continue;
+ }
+
+ Console.Write(choice.Message.Content[0].Text);
+ reply.Append(choice.Message.Content[0].Text);
+ usage = chunk.Usage;
+}
+
+Console.WriteLine();
+messages.Add(MultimodalMessage.Assistant([MultimodalMessageContent.TextContent(reply.ToString())]));
+if (usage != null)
+{
+ Console.WriteLine(
+ $"Usage: in({usage.InputTokens})/out({usage.OutputTokens})/image({usage.ImageTokens})/total({usage.TotalTokens})");
+}
+
+/*
+File uploaded: oss://dashscope-instant/52afe077fb4825c6d74411758cb1ab98/2025-11-28/435ea45f-9942-4fd4-983a-9ea8a3cd5ecb/tilted.jpg
+产品介绍
+本品采用韩国进口纤维丝制造，不缩水、不变形、不发霉、
+不生菌、不伤物品表面。具有真正的不粘油、吸水力强、耐水
+浸、清洗干净、无毒、无残留、易晾干等特点。
+店家使用经验：不锈钢、陶瓷制品、浴盆、整体浴室大部分是
+白色的光洁表面，用其他的抹布擦洗表面污渍不易洗掉，太尖
+的容易划出划痕。使用这个仿真丝瓜布，沾少量中性洗涤剂揉
+出泡沫，很容易把这些表面污渍擦洗干净。
+6941990612023
+货号：2023
+Usage: in(2434)/out(155)/image(2410)/total(2589)
+*/
+```
+
+
+
 ## 语音合成
 
 通过 `dashScopeClient.CreateSpeechSynthesizerSocketSessionAsync()` 来创建一个语音合成会话。

sample/Cnblogs.DashScope.Sample/Cnblogs.DashScope.Sample.csproj

@@ -29,6 +29,9 @@
  <None Update="sample.mp4">
  <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
  </None>
+ <None Update="tilted.png">
+ <CopyToOutputDirectory>Always</CopyToOutputDirectory>
+ </None>
  </ItemGroup>
 
  <ItemGroup>

sample/Cnblogs.DashScope.Sample/Multimodal/OcrSample.cs

@@ -0,0 +1,79 @@
+using System.Text;
+using Cnblogs.DashScope.Core;
+
+namespace Cnblogs.DashScope.Sample.Multimodal;
+
+public class OcrSample : ISample
+{
+ /// <inheritdoc />
+ public string Description => "OCR Sample with rotate enabled";
+
+ /// <inheritdoc />
+ public async Task RunAsync(IDashScopeClient client)
+ {
+ // upload file
+ await using var tilted = File.OpenRead("tilted.png");
+ var ossLink = await client.UploadTemporaryFileAsync("qwen-vl-ocr-latest", tilted, "tilted.jpg");
+ Console.WriteLine($"File uploaded: {ossLink}");
+ var messages = new List<MultimodalMessage>();
+ messages.Add(
+ MultimodalMessage.User(
+ [
+ MultimodalMessageContent.ImageContent(ossLink, enableRotate: true),
+ ]));
+ var completion = client.GetMultimodalGenerationStreamAsync(
+ new ModelRequest<MultimodalInput, IMultimodalParameters>()
+ {
+ Model = "qwen-vl-ocr-latest",
+ Input = new MultimodalInput() { Messages = messages },
+ Parameters = new MultimodalParameters()
+ {
+ IncrementalOutput = true,
+ }
+ });
+ var reply = new StringBuilder();
+ var first = false;
+ MultimodalTokenUsage? usage = null;
+ await foreach (var chunk in completion)
+ {
+ var choice = chunk.Output.Choices[0];
+ if (first)
+ {
+ first = false;
+ Console.Write("Assistant > ");
+ }
+
+ if (choice.Message.Content.Count == 0)
+ {
+ continue;
+ }
+
+ Console.Write(choice.Message.Content[0].Text);
+ reply.Append(choice.Message.Content[0].Text);
+ usage = chunk.Usage;
+ }
+
+ Console.WriteLine();
+ messages.Add(MultimodalMessage.Assistant([MultimodalMessageContent.TextContent(reply.ToString())]));
+ if (usage != null)
+ {
+ Console.WriteLine(
+ $"Usage: in({usage.InputTokens})/out({usage.OutputTokens})/image({usage.ImageTokens})/total({usage.TotalTokens})");
+ }
+ }
+}
+
+/*
+File uploaded: oss://dashscope-instant/52afe077fb4825c6d74411758cb1ab98/2025-11-28/435ea45f-9942-4fd4-983a-9ea8a3cd5ecb/tilted.jpg
+产品介绍
+本品采用韩国进口纤维丝制造，不缩水、不变形、不发霉、
+不生菌、不伤物品表面。具有真正的不粘油、吸水力强、耐水
+浸、清洗干净、无毒、无残留、易晾干等特点。
+店家使用经验：不锈钢、陶瓷制品、浴盆、整体浴室大部分是
+白色的光洁表面，用其他的抹布擦洗表面污渍不易洗掉，太尖
+的容易划出划痕。使用这个仿真丝瓜布，沾少量中性洗涤剂揉
+出泡沫，很容易把这些表面污渍擦洗干净。
+6941990612023
+货号：2023
+Usage: in(2434)/out(155)/image(2410)/total(2589)
+ */

sample/Cnblogs.DashScope.Sample/tilted.png

@@ -19,7 +19,8 @@ public record MultimodalMessageContent(
  int? MinPixels = null,
  int? MaxPixels = null,
  bool? EnableRotate = null,
- float? Fps = null)
+ float? Fps = null,
+ MultimodalOcrResult? OcrResult = null)
 {
  private const string OssSchema = "oss://";
 

src/Cnblogs.DashScope.Core/MultimodalMessageContent.cs

@@ -0,0 +1,20 @@
+namespace Cnblogs.DashScope.Core;
+
+/// <summary>
+/// Options for OCR model.
+/// </summary>
+public class MultimodalOcrOptions
+{
+ /// <summary>
+ /// Name of the task.
+ /// </summary>
+ /// <example>
+ /// Some task example: "text_recognition", "key_information_extraction", "document_parsing", "table_parsing"
+ /// </example>
+ public string? Task { get; set; }
+
+ /// <summary>
+ /// Config for the task.
+ /// </summary>
+ public MultimodalOcrTaskConfig? TaskConfig { get; set; }
+}

src/Cnblogs.DashScope.Core/MultimodalOcrOptions.cs

@@ -0,0 +1,8 @@
+namespace Cnblogs.DashScope.Core;
+
+/// <summary>
+/// OCR result from the model.
+/// </summary>
+/// <param name="WordsInfo">The words that model recognized.</param>
+/// <param name="KvResult">Meta info that extracted from the image.</param>
+public record MultimodalOcrResult(List<MultimodalOcrWordInfo>? WordsInfo, Dictionary<string, object?> KvResult);

src/Cnblogs.DashScope.Core/MultimodalOcrResult.cs

@@ -0,0 +1,28 @@
+namespace Cnblogs.DashScope.Core;
+
+/// <summary>
+/// Configuration of OCR task.
+/// </summary>
+public class MultimodalOcrTaskConfig
+{
+ /// <summary>
+ /// The resulting JSON schema, value should be empty string.
+ /// </summary>
+ /// <example>
+ /// <code>
+ /// var schema = new Dictionary&lt;string, object&gt;()
+ /// {
+ /// {
+ /// "收件人信息",
+ /// new Dictionary&lt;string, object&gt;()
+ /// {
+ /// "收件人姓名", string.Empty,
+ /// "收件人电话号码", string.Empty,
+ /// "收件人地址", string.Empty
+ /// }
+ /// }
+ /// }
+ /// </code>
+ /// </example>
+ public Dictionary<string, object>? ResultSchema { get; set; }
+}

src/Cnblogs.DashScope.Core/MultimodalOcrTaskConfig.cs

@@ -0,0 +1,9 @@
+namespace Cnblogs.DashScope.Core;
+
+/// <summary>
+/// Word location and info recognized by the model.
+/// </summary>
+/// <param name="Text">OCR result.</param>
+/// <param name="RotateRect">Four point of the word rect, (0, 0) is the left-top of the image. The points are clockwise, starting from the left-top of the rect. e.g. [x1, y1, x2, y2, x3, y3, x4, y4]</param>
+/// <param name="Location">Another presentation of the word rect. First two are the center point of the rect, then follows the width and height. The last value is rect's rotate angle from the landscape. e.g. [center_x, center_y, width, height, angle]</param>
+public record MultimodalOcrWordInfo(string Text, int[] RotateRect, int[] Location);