task_three_d__object_cluster__total_object_count

Summary

• Domain: three_d
• Scene id: object_cluster
• Package: src/trace_tasks/tasks/three_d/object_cluster/
• Supported query_id: single
• Answer type: integer
• Annotation type: unordered bbox_set
• Annotation schema: bbox_set

Program Contract

Program: count(filter(object_cluster_objects, is_countable_object = true)); scene=object_cluster; scope=total_object_count

Candidate set: the visible 3D objects, surfaces, room/street/warehouse structures, spatial anchors, markers, and labeled options inside the total_object_count objective scope. Operands: visible scene state and prompt-bound operands named by filter, object_cluster_objects, is_countable_object, true, object_cluster, total_object_count. 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.

Reasoning Operations

Families: counting

Contract

The image shows many small synthetic perspective 3D objects arranged on a plain surface. The objects remain visually clustered, but generation may use one or more local centers with continuous compactness variation so scenes range from more compact to more spread out. This level-0 cluster task asks for the total number of visible objects, without filtering by object type, color, relation, or region.

Generation uses a one-type object set: all visible objects are countable instances from one sampled color-safe object type. The sampled object type, cluster count, compactness, and object colors are render variety metadata only and are not named in the prompt. There are no unrelated distractor objects in this task. The default generated object-count support is 6-20.

Visible objects use 2-4 canonical named colors for non-semantic visual variety when the count permits. The color distribution is recorded in trace metadata as color_role = non_semantic_visual_variation, visual_color_names, and visual_color_counts; color is not part of the answer predicate.

The answer is the integer count of finalized visible objects with is_countable_object = true. Pixels are render output, not verifier source of truth.

Annotation Contract

Annotation is a bbox_set containing one [x0, y0, x1, y1] pixel box around each counted object. The annotation set is unordered because all witnesses have the same semantic role and annotation cardinality matches the answer.

Prompt And Trace

The prompt bundle is three_d_object_cluster_v1 under src/trace_tasks/resources/prompts/three_d/object_cluster/. The trace records camera pose, projection frame, object world coordinates, sampled dimensions, primary object type metadata, cluster layout metadata, composition offset metadata, non-semantic color distribution, rendered readability stats, all counted object ids, projected object boxes, and the 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 3D scene trace.