Showcase Viewer

WorldAtlas includes a lightweight showcase viewer for inspecting generated media before or after scoring. It supports three visualization workflows: multi-view browsing, method-to-method comparison, and GIF/image comparison.

The viewer is static. It scans prediction folders, writes showcase_manifest.json, links media into a local media/ folder, and renders index.html. No Dash, Gradio, or remote service is required.

Method Comparison

Compare two or more WorldAtlas prediction roots:

python scripts/showcase/video_method_compare_app.py \
  --method astra=artifacts/predictions/astra \
  --method yume=artifacts/predictions/yume \
  --output-dir .cache/showcase/astra_yume

Open .cache/showcase/astra_yume/index.html in a browser, or add --serve --port 7860 to serve it locally.

Multi-View Browsing

For folders such as generated_results/<method>/<sample>/<camera>.mp4, run:

python scripts/showcase/video_multi_view_app.py \
  --generated-results generated_results \
  --output-dir .cache/showcase/multiview

The scanner groups common camera names such as CAM_FRONT, CAM_BACK, front_left, and right into the same sample row.

GIF/Image Comparison

For GIFs or image previews:

python scripts/showcase/gif_method_compare_app.py \
  --generated-results generated_results \
  --output-dir .cache/showcase/gifs

This mode filters out video-only files and keeps .gif, .png, .jpg, .jpeg, and .webp.

WorldAtlas Generation Records

If a method root contains generation_records*.jsonl, the viewer uses the recorded sample_id, suite, prediction_path, and prompt fields instead of inferring sample names from filenames. Add --include-conditioning to show conditioning inputs next to predictions.

Reference media can be included when a benchmark manifest is available:

worldatlas showcase \
  --method astra=artifacts/predictions/astra \
  --benchmark-manifest artifacts/benchmark/manifest/worldatlas_manifest.jsonl \
  --include-reference \
  --output-dir .cache/showcase/astra_with_reference