본문으로 건너뛰기

Plugin Categories

Plugin categories organize plugins by their primary functionality.

PluginCategory Enum​

from synapse_sdk.plugins.enums import PluginCategory
CategoryValueDescription
NEURAL_NET'neural_net'Machine learning model operations including training, inference, and deployment
EXPORT'export'Data export and transformation operations
UPLOAD'upload'File and data upload functionality
SMART_TOOL'smart_tool'Intelligent automation tools and utilities
POST_ANNOTATION'post_annotation'Post-processing workflows after data annotation
PRE_ANNOTATION'pre_annotation'Pre-processing workflows before data annotation
DATA_VALIDATION'data_validation'Data quality checks and validation operations
CUSTOM'custom'Custom plugins that don't fit other categories

Usage​

In config.yaml​

name: My Training Plugin
code: my-training-plugin
version: 1.0.0
category: neural_net

actions:
  train:
    entrypoint: plugin.train.TrainAction
    method: job

In Python Code​

from synapse_sdk.plugins.enums import PluginCategory
from synapse_sdk.plugins.action import BaseAction

class TrainAction(BaseAction):
    category = PluginCategory.NEURAL_NET
    action_name = 'train'

    def execute(self):
        return {'status': 'completed'}

With @action Decorator​

from synapse_sdk.plugins.decorators import action
from synapse_sdk.plugins.enums import PluginCategory

@action(name='process', category=PluginCategory.SMART_TOOL)
def process_data(params, ctx):
    return {'processed': True}

Category-Specific Features​

NEURAL_NET​

Plugins for machine learning workflows.

Common actions:

  • train - Model training
  • inference - Model inference/prediction
  • export - Model export to various formats
  • tune - Hyperparameter tuning
  • deployment - Model deployment

Example:

name: YOLOv8 Plugin
code: yolov8
version: 1.0.0
category: neural_net

actions:
  train:
    entrypoint: plugin.train.TrainAction
    method: job
    description: Train YOLOv8 model

  infer:
    entrypoint: plugin.inference.InferAction
    method: task
    description: Run inference on images

SMART_TOOL​

Interactive and automatic annotation tools.

Common actions:

  • annotate - Automatic annotation
  • suggest - Annotation suggestions
  • validate - Annotation validation
from synapse_sdk.plugins.enums import SmartToolType, AnnotationType

class AutoAnnotateAction(BaseAction):
    category = PluginCategory.SMART_TOOL
    tool_type = SmartToolType.AUTOMATIC
    annotation_type = AnnotationType.BBOX

DATA_VALIDATION​

Data quality and validation plugins.

Common actions:

  • validate - Validate data against rules
  • check - Run quality checks
  • report - Generate validation reports

RunMethod​

Execution methods for plugin actions.

from synapse_sdk.plugins.enums import RunMethod

RunMethod.JOB    # Async execution via Ray Job API
RunMethod.TASK   # Fast execution via Ray Actor
RunMethod.SERVE  # Ray Serve deployment

DataType​

Data types handled by plugins.

from synapse_sdk.plugins.enums import DataType

DataType.IMAGE   # Image data
DataType.TEXT    # Text data
DataType.VIDEO   # Video data
DataType.PCD     # Point cloud data
DataType.AUDIO   # Audio data

AnnotationCategory​

Annotation categories for smart tools.

from synapse_sdk.plugins.enums import AnnotationCategory

AnnotationCategory.IMAGE    # Image
AnnotationCategory.VIDEO    # Video
AnnotationCategory.AUDIO    # Audio
AnnotationCategory.TEXT     # Text
AnnotationCategory.PCD      # PCD
AnnotationCategory.PROMPT   # Prompt

AnnotationType​

Annotation types for smart tools.

from synapse_sdk.plugins.enums import AnnotationType

AnnotationType.BBOX     # Bounding box
AnnotationType.POLYGON  # Polygon
AnnotationType.POINT    # Point
AnnotationType.LINE     # Line/polyline
AnnotationType.MASK     # Segmentation mask
AnnotationType.LABEL    # Classification label

SmartToolType​

Smart tool implementation types.

from synapse_sdk.plugins.enums import SmartToolType

SmartToolType.INTERACTIVE      # User-interactive tool
SmartToolType.AUTOMATIC        # Fully automatic
SmartToolType.SEMI_AUTOMATIC   # Semi-automatic with user input