\n",
"
\n",
" ------SYSTEM_PROMPT------\n",
"
\n",
" You are working on a 3D visual grounding task, which involves receiving a query that specifies a particular object by describing its attributes and grounding conditions to uniquely identify the object. Here, attributes refer to the inherent properties of the object, such as category, color, appearance, function, etc. Grounding conditions refer to considerations of other objects or other conditions in the scene, such as location, relative position to other objects, etc.\n",
"Now, I need you to first parse this query, return the category of the object to be found, and list each of the object's attributes and grounding conditions. Each attribute and condition should be returned individually. Sometimes the object's category is not explicitly specified, and you need to deduce it through reasoning. If you cannot deduce after reasoning, you can use 'unknown' for the category.\n",
"Your response should be formatted using JSON contained in ```json```.\n",
"\n",
"Here are some examples:\n",
"\n",
"Input:\n",
"Query: this is a brown cabinet. it is to the right of a picture.\n",
"\n",
"Output:\n",
"```\n",
"{\n",
"\"target_class\": \"cabinet\",\n",
"\"attributes\": [\"it's brown\"],\n",
"\"conditions\": [\"it's to the right of a picture\"]\n",
"}\n",
"```\n",
"\n",
"Input:\n",
"Query: it is a wooden computer desk. the desk is in the sleeping area, across from the living room. the desk is in the corner of the room, between the nightstand and where the shelf and window are.\n",
"\n",
"Output:\n",
"```\n",
"{\n",
"\"target_class\": \"desk\",\n",
"\"attributes\": [\"it's a wooden computer desk\"],\n",
"\"conditions\": [\"it's in the sleeping area\", \"it's across from the living room\", \"it's in the corner of the room\", \"it's between the nightstand and where the shelf and window are\"]\n",
"}\n",
"```\n",
"\n",
"Input:\n",
"Query: In the room is a set of desks along a wall with windows totaling 4 desks. Opposite this wall is another wall with a door and two desks. The desk of interest is the closest desk to the door. This desk has nothing on it, no monitor, etc.\n",
"\n",
"Output:\n",
"```\n",
"{\n",
"\"target_class\": \"desk\",\n",
"\"attributes\": [], # no attributes\n",
"\"conditions\": [\"it's near the wall with a door and two desks, opposite a wall with windows totaling four desks\", \"it's closest to the door on the wall\", \"it has nothing on it, no monitor, etc\"] # need some reasoning to write the conditions\n",
"}\n",
"```\n",
"Ensure your response adheres strictly to this JSON format, as it will be directly parsed and used.\n",
"
\n",
"------END_SYSTEM_PROMPT------\n",
"\n",
"\n",
"this is a white keyboard. it is located below the computer screen on the right.\n",
"\n",
"
\n",
"
\n",
" {\n",
"\"target_class\": \"keyboard\",\n",
"\"attributes\": [\"it's white\"],\n",
"\"conditions\": [\"it's located below the computer screen on the right\"]\n",
"}\n",
"\n",
"
\n",
"