Spectral-Spatial Interpretation attributes images by position and channel
Original title: Spectral-Spatial Interpretation
On HLS Burn Scar, SSI beat an MC-Owen estimator while using about 6% of its evaluation budget.
Researchers introduce Spectral-Spatial Interpretation, a full-grid attribution method for multispectral images that computes Shapley values over channels at each spatial position. The result is a spectral-spatial matrix rather than one heatmap or a global channel ranking. Across CUB-200, EuroSAT, and HLS Burn Scar, the authors report the best joint deletion and insertion AUC and the strongest spectral hit rates. Code is public.
Unconfirmed: Metric gains are the authors' reported results; independent replication is not stated.
- Channel attributions at a position recover that position's contextual contrast.
- It differs from context-averaged Owen-style hierarchical accounting.
- Tests used per-channel mean and zero baselines.
- Code is at github.com/AtheavenMa/Spectral-Spatial-Interpretation.
From the original article
1 min read at huggingface.co
Post-hoc image explanations usually show where a model looks using spatial heatmaps or indicate which channels matter using global channel scores. These separate views are limiting for multispectral data because channel relevance is often spatially localized and physically meaningful. We introduce Spectral-Spatial Interpretation (SSI), a full-grid attribution method over position-channel players (g,c).
By Haotian Ma, Ruqi Yang, Philip Townsend · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.