张量转图片 (WriteTensorToImage)

Java 类名：com.alibaba.alink.pipeline.image.WriteTensorToImage

Python 类名：WriteTensorToImage

功能介绍

将张量列转换为图片，并写入根目录对应的相对路径列中，然后原样输出结果。

参数说明

名称	中文名称	描述	类型	是否必须？	取值范围	默认值
relativeFilePathCol	文件路径列	文件路径列	String	✓
rootFilePath	文件路径	文件路径	String	✓
tensorCol	tensor列	tensor列	String	✓
imageType	图片类型	图片类型	String		“PNG”, “JPEG”	“PNG”
reservedCols	算法保留列名	算法保留列	String[]			null

代码示例

Python 代码

df_data = pd.DataFrame([
    'sphx_glr_plot_scripted_tensor_transforms_001.png'
])

batch_data = BatchOperator.fromDataframe(df_data, schemaStr = 'path string')

readImageToTensorBatchOp = ReadImageToTensorBatchOp()\
    .setRootFilePath("https://pytorch.org/vision/stable/_images/")\
	.setRelativeFilePathCol("path")\
	.setOutputCol("tensor")

writeTensorToImageBatchOp = WriteTensorToImageBatchOp()\
			.setRootFilePath("/tmp/write_tensor_to_image")\
			.setTensorCol("tensor")\
			.setImageType("png")\
			.setRelativeFilePathCol("path")

batch_data.link(readImageToTensorBatchOp).link(writeTensorToImageBatchOp).print()

Java 代码

import org.apache.flink.types.Row;

import com.alibaba.alink.operator.batch.source.MemSourceBatchOp;
import com.alibaba.alink.params.image.HasImageType.ImageType;
import com.alibaba.alink.pipeline.image.WriteTensorToImage;
import org.junit.Test;

import java.util.Collections;
import java.util.List;

public class WriteTensorToImageTest {

	@Test
	public void testWriteTensorToImage() throws Exception {

		List <Row> data = Collections.singletonList(
			Row.of("sphx_glr_plot_scripted_tensor_transforms_001.png")
		);

		MemSourceBatchOp memSourceBatchOp = new MemSourceBatchOp(data, "path string");

		ReadImageToTensorBatchOp readImageToTensorBatchOp = new ReadImageToTensorBatchOp()
			.setRootFilePath("https://pytorch.org/vision/stable/_images/")
			.setRelativeFilePathCol("path")
			.setOutputCol("tensor");

		WriteTensorToImage writeTensorToImageBatchOp = new WriteTensorToImage()
			.setRootFilePath("/tmp/write_tensor_to_image")
			.setTensorCol("tensor")
			.setImageType(ImageType.PNG)
			.setRelativeFilePathCol("path");

		writeTensorToImageBatchOp.transform(memSourceBatchOp.link(readImageToTensorBatchOp)).print();
	}

}

运行结果

可以在 /tmp/write_tensor_to_image/sphx_glr_plot_scripted_tensor_transforms_001.png 中找到 https://pytorch.org/vision/stable/_images/sphx_glr_plot_scripted_tensor_transforms_001.png

同时组件的输出结果为：