LVM image editing algorithms provide features such as image cleaning, content filtering, basic image information extraction, and image caption generation. You can combine different algorithms to filter image data and generate text descriptions. This process provides high-quality image data for training image generation models. This topic describes how to use the image-text filtering preset template in Machine Learning Designer.
Limits
The image-text filtering preset template is available only in the China (Hangzhou), China (Shanghai), China (Beijing), and China (Shenzhen) regions.
Prepare image data
PAI provides sample data that you can use:
Download the image metadata file and the image file.
Image metadata file: image_meta.jsonl. This file is used as the input for image-text algorithms.
Image file: data.zip. This file is used as the input for general image editing algorithms.
Decompress the package and upload the image files to OSS. For more information, see Simple upload.
Modify the image metadata file.
Replace the configuration
oss://bucket_name.oss-cn-hangzhou.aliyuncs.com/image_algorithm_test/image_data/in the image metadata file with the OSS directory where you uploaded the images.
Upload the image metadata file to the same OSS bucket where you uploaded the image files. For more information, see Simple upload.
Create and run a pipeline
Go to the Machine Learning Designer page.
Log on to the PAI console.
In the upper-left corner, select the desired region.
In the navigation pane on the left, click Workspaces, and then click the name of the workspace that you want to manage.
In the navigation pane on the left, choose Model Training > Visualized Modeling (Designer).
Create a pipeline.
On the Preset Templates tab, choose Business Area > Multi-modal LLM, and then click Create in the Image-Text Filtering template.
Configure the pipeline parameters and click OK. You can use the default values.
In the pipeline list, select the pipeline that you created and click Open.
Configure the pipeline.
Configure the Read File Data component. Click the Read File Data component. On the Field Settings tab in the right-side pane, set OSS Data Path to the OSS directory where the image data file is stored.
Configure the LLMDataProcessGroup1 group. Click the Settings button
and configure Data Output OSS Directory. The output file is saved to this directory. For more information about the configuration of the LVM image pre-processing algorithm component, see Image preprocessing operators.
Run the pipeline. After the run is complete, you can view the generated files:
meta.jsonl file: During the run, the meta.jsonl image metadata file is generated in the parent directory of the path specified for OSS Path of Image Data.
Result file: You can view the result file in the directory that you specified for OSS Path of Output File.
For more information about the result file, see the description of the OSS Path of Output File parameter in Image preprocessing operators.