LVM-Image-Shape Filter (DLC)-Platform For AI(PAI)-阿里云帮助中心

The LVM-Image-Shape Filter (DLC) component filters images by resolution, discarding any image whose width or height falls outside the specified pixel range. Use this component to clean training data before running image generation model training jobs.

Supported computing resources

Deep Learning Containers (DLC)

How it works

The component reads image data from Object Storage Service (OSS), calculates the width and height of each image in pixels, and removes any image whose dimensions fall outside the configured minimum-maximum range. Images that pass all threshold checks are written to the output path as a filtered dataset. In most cases, the component is used for the subsequent training of image generation models.

The component generates an image metadata file (meta.jsonl) in the parent directory of the path you specify in Image Data OSS Path. On subsequent runs, point directly to this file instead of the original directory to skip re-scanning the source data.

Inputs and outputs

Input ports

Use one of the following as the upstream input:

Read File Data — reads the OSS path where your training data is stored.
Image Data OSS Path — specify the OSS directory containing your images, or point directly to a previously generated meta.jsonl metadata file.
Any image data preprocessing component.

Output port

The filtered dataset. The output path contains three files:

File	Description
`{name}.jsonl`	The filtered output file. The filename is set by the Output Filename parameter (default: `result.jsonl`).
`{name}_stats.jsonl`	The state file.
`dj_run_yaml.yaml`	The parameter configuration snapshot used when the algorithm ran.

Configure the component

Configure the LVM-Image-Shape Filter (DLC) component in Machine Learning Designer. The following table describes all parameters.

Tab	Parameter	Required	Default	Description
Field Settings	Image Data OSS Path	No	—	The OSS directory containing your image data, or the path to a previously generated `meta.jsonl` file. On the first run (when no upstream component exists), specify the directory manually. The component writes `meta.jsonl` to the parent directory of this path.
Field Settings	OSS Directory for Saving OutputData	No	—	The OSS directory where the filtering results are stored.
Field Settings	Output Filename	No	`result.jsonl`	The filename of the output file (without the `_stats` suffix).
Parameter Settings	Minimum Width	Yes	`1`	The minimum image width, in pixels. Images narrower than this value are filtered out.
Parameter Settings	Maximum Width	Yes	`2048`	The maximum image width, in pixels. Images wider than this value are filtered out.
Parameter Settings	Minimum Height	Yes	`1`	The minimum image height, in pixels. Images shorter than this value are filtered out.
Parameter Settings	Maximum Height	Yes	`2048`	The maximum image height, in pixels. Images taller than this value are filtered out.
Execution Tuning	Number of Processes	No	`4`	The number of parallel processes to use during filtering.
	Single-node/Multi-node	No	single-node	Single-node or multi-node mode. Default value: single-node. Multi-node mode supports only the Ray framework.
	Select Resource Group	No	—	Select a public resource group or a resource quota (if distributed mode is used, you must set resource parameters for the head and worker nodes separately): Public resource group: We recommend that you use the CPU instance type to save costs. Resource quota: select the number of GPUs, number of vCPUs, memory (GB), and shared memory (GB).
	Maximum Running Duration (seconds)	No	—	The time limit for the job, in seconds. The job is terminated if it exceeds this limit.