AffordanceBench Studio

Open, software-first toolkit for query processing, evaluation, visualization, profiling, and reproducibility workflows in 3D multimedia affordance research.

Repository Release v0.1.2 Source ZIP Overview PDF Supplementary ZIP

Why This Page Exists

This page is designed for OSS reviewers and external users who want one place to inspect the toolkit, download the archive, preview the demo, and verify buildability evidence without digging through a research-only workspace.

Software identity Unified CLI + public release
Buildability signals Linux smoke CI + source-ZIP proof
Demo readiness GIF / MP4 + output gallery
Real evidence OpenAD smoke + LASO heatmap smoke

One-Minute Reviewer Path

pip install -r requirements.txt
pip install -e .

affordbench env-check
affordbench list
affordbench describe laso-qaq
bash examples/demo_simulation_reviewer_walkthrough.sh

This path emphasizes environment diagnostics, command discovery, dry-run resolution, and pointers to validated remote smoke evidence.

Demo And Output Gallery

Simulation-First Demo

Simulation-first reviewer demo

Lightweight reviewer walkthrough covering environment checks, command discovery, dry-run resolution, and public evidence links.

Reviewer Output Gallery

Reviewer-facing output gallery

Combines environment inspection, dry-run cards, a real anchor-map JSON preview, a backup-manifest preview, heatmap evidence, and a profiling summary card.

Validation Evidence

Install And Packaging

Execution Evidence

Software Snapshot

What Reviewers Can Inspect Quickly

  • Setup: `env-check`, `list`, `describe`, and CI-backed smoke
  • Workflow coverage: training, evaluation, LASO, visualization, profiling, and packaging helpers
  • Compatibility: explicit env vars, runtime shims, `openai_CLIP` auto-discovery, CPU-safe smoke path
  • Packaging maturity: `README.md`, `pyproject.toml`, `requirements.txt`, `Dockerfile`, `docs/`, and `submission/`

Documentation

Public-Safe Scope

This release is intentionally scoped for software inspection. It does not bundle large benchmark datasets, under-review main-paper checkpoints, or final paper-facing figures by default.