Skip to content

task_three_d__carousel__belt_total_object_count

Summary

  • Domain: three_d
  • Scene id: carousel
  • Scene package: carousel
  • Query ids: single publicly; internal query id belt_total_count
  • Answer type: integer
  • Annotation type: unordered bbox_set
  • Annotation schema: bbox_set

Program Contract

Program: count(filter(conveyor_objects, belt_key=target_belt_key)); scene=carousel; scope=belt_total_object_count

Candidate set: the visible 3D objects, surfaces, room/street/warehouse structures, spatial anchors, markers, and labeled options inside the belt_total_object_count objective scope. Operands: visible scene state and prompt-bound operands named by filter, conveyor_objects, belt_key, target_belt_key, carousel, belt_total_object_count plus the active query_id branch. Operation: evaluate count over the candidate set using the finalized 3D scene state, camera projection, object identities, spatial relations, counts, distances, or option-selection constraints encoded in the program expression; generation enforces a unique final answer. Output binding: answer uses the integer schema; generation binds a unique final answer. Annotation witnesses: annotation uses the bbox_set schema; the prompt/annotation contract defines the minimal visual witnesses. Query ids: single, belt_total_count.

Reasoning Operations

Families: counting

Contract

The image shows one 3D conveyor carousel with two visible concentric elliptical belts: an inner belt and an outer belt. The belts are distinguished by position, not by text written on the image. Small 3D objects sit on the belt surfaces.

Each generated instance uses one sampled object type across both belts. Object colors may vary, but color is not part of the task predicate. The task asks for the total number of visible objects on the requested belt.

The target belt is sampled as a generation axis. If the target belt is inner, the answer support is 1..8. If the target belt is outer, the answer support is 1..10.

Annotation Contract

Annotation is a bbox_set containing one [x0, y0, x1, y1] pixel box around each counted object on the requested belt. Objects on the other belt, belt surfaces, arrows, and decorative station context are not annotation.

Prompt And Trace

The prompt bundle is three_d_carousel_v1 under src/trace_tasks/resources/prompts/three_d/carousel/. The prompt asks for total objects on the INNER or OUTER belt and does not mention the sampled object type or colors.

The trace records scene variant, belt records, target belt, sampled object type, object specs, projected object boxes and centers, camera, projection frame, and solver count predicate.

Determinism

Generation is deterministic from instance_seed, explicit params, config defaults, prompt bundle, and code versions. Answers and annotation come from the same finalized conveyor trace.