Polo123/NASA-Crater-detection

NASA Crater Detection - Interactive Inference Space

This Space is built for inference-only crater detection with ONNX Runtime and includes technical visualizations of the decode pipeline.

What this Space shows

• —Downloadable sample images (plus zip) for quick testing
• —Final crater ellipses overlaid on the uploaded image
• —Heatmap visualization
• —Local maxima after thresholding
• —Top-K candidate centers
• —Kept points after adaptive NMS
• —Full candidate table with rejection reasons
• —Downloadable contest-format CSV

Model pipeline (high-level)

1. 1.Predict heatmap, size (a, b), angle terms (sin2theta, cos2theta), and offsets.
2. 2.Detect candidate peaks by threshold + local maxima.
3. 3.Decode center coordinates and ellipse parameters.
4. 4.Apply Top-K prefilter.
5. 5.Apply adaptive NMS.
6. 6.Keep valid detections and export final results.

Files

• —app.py - Gradio UI
• —space_inference.py - ONNX inference + decode logic
• —visualization.py - technical rendering helpers
• —model/model.onnx + model/model.onnx.data - ONNX graph and external weights
• —assets/samples/ - 3 train images for download + one zip bundle

Local run

pip install -r requirements.txt
python app.py