LVM-Tag Mapper (DLC) is a Platform for AI (PAI) component that uses the RAM (Recognize Anything Model) algorithm to calculate the tags of sampled frames from MP4 videos. The tagged output is stored as JSONL files in Object Storage Service (OSS) for use in downstream large vision model (LVM) data processing pipelines.
How it works
-
The component samples frames from each MP4 video at evenly spaced intervals based on video duration.
-
The RAM algorithm processes each sampled frame and calculates tags.
-
The tagged results are written to a JSONL file in the specified OSS output directory.
-
A parameter configuration file (
dj_run_yaml.yaml) is also saved alongside the results.
Prerequisites
Before you begin, ensure that you have:
-
An OSS bucket containing MP4 video files or a video metadata file (
video_meta.jsonl) -
Access to Machine Learning Designer in PAI
-
A GPU instance type available in a public or dedicated resource group
This component requires a GPU instance type. CPU instance types are not supported.
Supported computing resources
Inputs and outputs
Input ports
The component accepts the following input sources:
-
Read File Data component: Reads the OSS path where video data is stored.
-
Any LVM Data Processing (DLC) component: Connect any upstream LVM Data Processing component directly.
To specify the video source, configure the OSS Data Path parameter to select the OSS directory containing MP4 video files, or select the video metadata file (video_meta.jsonl). You can also set Video Data OSS Path directly in the parameter table below.
On the first run without an upstream component, select the OSS directory manually using Video Data OSS Path. The component generates video_meta.jsonl in the parent directory of the specified path. On subsequent runs, select video_meta.jsonl directly to skip re-scanning the directory.
Output port
The component writes two files to the OSS directory specified by Output File OSS Path:
| File | Description |
|---|---|
{name}.jsonl |
The tag results for each sampled frame. The filename is set by the Output Filename parameter (default: result.jsonl). Pass this file to downstream LVM Data Processing components. |
dj_run_yaml.yaml |
The algorithm parameter configuration used during the run. Use this file to audit or reproduce the run configuration. |
Configure the LVM-Tag Mapper (DLC) component
Configure the LVM-Tag Mapper (DLC) component parameters in Machine Learning Designer.
Field settings
| Parameter | Required | Description | Default |
|---|---|---|---|
| Video Data OSS Path | No | The OSS directory containing the video files, or the video_meta.jsonl file from a previous run. Required if no upstream component is connected on the first run. |
None |
| Output File OSS Path | Yes | The OSS directory where output files are saved. | None |
| Output Filename | Yes | The filename for the tag results file. The file is saved in JSONL format. | result.jsonl |
Parameter settings
| Parameter | Required | Description | Default |
|---|---|---|---|
| Number of sampled frames | Yes | The number of frames sampled per video. Frames are selected at evenly spaced intervals based on video duration. | 3 |
Execution tuning
| Parameter | Required | Description | Default |
|---|---|---|---|
| Number of processes | Yes | The number of parallel processes for tag computation. | 4 |
Select resource group
| Option | Required | Description | Default |
|---|---|---|---|
| Public resource group | No | Select the GPU instance type and VPC. Must be a GPU instance type. | None |
| Dedicated resource group | No | Specify the number of vCPUs, memory, shared memory, and GPUs. | None |
| Maximum running duration (seconds) | No | The maximum time the component is allowed to run. The job is terminated if this limit is exceeded. | None |