CoolFace
Apppublic

ozgurcuk/HorusAtlasDemo

sourceHugging Facemitupdated 5mo agoView on Hugging Face
0likes
App README

HorusAtlas Aerial Analyst Demo

This Hugging Face Space is a cache-first product demo for HorusAtlas.

It does not run full live inference by default. The Space loads validated local artifacts:

  • —aerial showcase cases with structured routing,
  • —a guided presentation mode for the strongest story beats,
  • —a validation-only Commons holdout to reduce cherry-picking risk,
  • —structured routing and readiness gates,
  • —a curated xBD disaster showcase locked for a stable demo,
  • —a live-test upload path when a HorusAtlas API backend is configured.

Live testing new images

The public Space stays light and cache-first, but the upload tab can test new images if you connect it to a running HorusAtlas backend:

bash
export HORUSATLAS_API_URL="https://YOUR-HORUSATLAS-BACKEND"
# optional if your backend is protected
export HORUSATLAS_API_TOKEN="..."

The backend must expose POST /v1/image-commentary.

From the HorusAtlas repository root, a local backend can be started with:

bash
tools/run_live_backend.sh

Experimental in-Space live inference is also opt-in:

bash
export HORUSATLAS_ENABLE_LIVE=1

Use in-Space live inference only on a GPU Space or an environment with the full HorusAtlas runtime and model cache.

Why cache-first

Full live inference is too slow for a reliable CPU-only Space. This demo is designed for presentation stability. Use HORUSATLAS_API_URL for practical live upload testing, or a GPU Space with the full runtime later.

Recommended demo flow

  1. 1.Open Presentation Mode and show the solar farm correction.
  2. 2.Move through quarry and stadium to show transparent routing.
  3. 3.Open xBD Disaster Showcase to show the locked catastrophe path.
  4. 4.End with Upload Preview or Live Upload if HORUSATLAS_API_URL is configured.

Publish

From this folder:

bash
export HF_SPACE_ID="YOUR_USERNAME/horusatlas-aerial-demo"
./publish_hf_space.sh

If hf is missing, install the Hugging Face CLI first, then authenticate with a token from your Hugging Face account settings.