Question: [question]

# Scene Graph Update Request

## Current Scene Graph Information
[relation_subgraph]

## Input Text
The following text describes a new Q&A with VLM within the period '{period}':
Q: {llm_instruction}
A: {vlm_response}

## Task
Based on the input text, extract new objects, relationships, and actions that should be added to the current scene graph for this period.

1. For objects: Compare with the existing objects and identify any new objects mentioned in the input text.
2. For relationships: Identify static relationships (not actions) between entities, such as spatial relationships, containment, or attributes.
3. For actions: Identify dynamic interactions where a person performs an action on or with other entities.
Only add new information related to the question.

## Output Format
```json
{
    "new_entities": [
        {
            "name": "object_name",
            "info": "description of the object"
        }
    ],
    "new_relationships": [
        {
            "relation_name": "relationship_type",
            "source_name": "source_entity_name", // no person
            "target_name": "target_entity_name", // no person
            "relation_info": "description of the relationship",
            "start_time": "hh:mm:ss",
            "end_time": "hh:mm:ss" // optional
        }
    ],
    "new_actions": [
        {
            "action_name": "action_verb",
            "action_info": "description of the action",
            "time_range": "{period}",
            "agent": "person_name",
            "patient": "affected_entity_name",  // optional
            "instrument": "tool_entity_name",   // optional
            "source": "starting_location",      // optional
            "target": "ending_location"         // optional
            "prev_action_id": "previous_action_id" // required, make sure the sequence of actions is correct. format: {agent}_{action_name}_{patient}_{time_range}
        }
    ]
}
```
