WorldFoundry
English中文

SpatialReasoner

SpatialReasoner from Johns Hopkins University (CCVL): multimodal-reasoning and spatial-reasoning; Runner parity pending.

Edit this pageGitHubMarkdown

SpatialReasoner is a vision-language model from the JHU CCVL group focused on explicit, step-wise spatial reasoning over images and video. In this catalog it is an understanding entry — spatial and multimodal reasoning plus image/video QA — returning model_response artifacts.

No official teaser or method figure is checked into this docs tree.

Code: ccvl/SpatialReasoner.

Build your run

Pipeline
SpatialReasonerPipelineworldfoundry.pipelines.video_official.pipeline_official_video
Task
Environmentworldfoundry-unified-cu128
DeviceCUDA 12.8
Runnerworldfoundry.pipeline
Loadingfrom_pretrained
Invocationnative_or_standard_invocation
worldfoundry-eval run \
  spatial-reasoner \
  --pipeline.task-profile multimodal-reasoning \
  --json

What it is for

Use spatial-reasoner for multimodal reasoning inference on this card.

A vision-language model with explicit 3D-aware spatial reasoning stages (per the upstream release at ccvl/SpatialReasoner). This repository records the prompt/image/video contract; scale and training details are not recorded here. It is published by Johns Hopkins University (CCVL).

Catalog tasks include multimodal-reasoning, spatial-reasoning, image-question-answering, and video-question-answering. Catalog aliases include spatial_reasoner and ccvl/SpatialReasoner.

Runtime contract

Optional prompt, image, video, seed, and frames; artifacts spatial-reasoner.json.

CLI task profiles: multimodal-reasoning, spatial-reasoning, image-question-answering, video-question-answering, and video_generation. WorldFoundry binds SpatialReasonerPipeline.

Family status: Runner parity pending.

Pair it with generation entries to audit spatial consistency of generated worlds — e.g. object positions and relative depths.

WorldFoundry route is complete; missing checkpoint/assets are reported at runtime.

Running in WorldFoundry

Launch worldfoundry-eval run spatial-reasoner --pipeline.task-profile multimodal-reasoning to run the reasoning route. The command builder above switches --pipeline.task-profile among recorded tasks and still emits worldfoundry-eval run.

It runs in the unified environment worldfoundry-unified-cu128 (Python 3.11, CUDA 12.8).

Recorded weight repositories include ccvl/SpatialReasoner. Prepare the environment and check local assets with bash scripts/inference/prepare_model_infer.sh spatial-reasoner. Add --download to fetch missing public checkpoints.

Status: Runner parity pending.