How iauh23/spectrum-fusion works
spectrum-fusion is a Python evaluation harness for comparing RGB and thermal person detectors against point annotations. A user runs the baseline entry point, which loads ground truth and processes each modality’s images through detection, point matching, and frame-weighted/localization scoring, then saves per-frame and aggregate results. RGB/thermal fusion, overlap analysis, and confidence sweeps are documented future work, not implemented runtime stages. The source excerpts show the main calls; dataset pairing is also documented in the README.
Main entry points: baseline.py Baseline Runner · gt_parser.py GT Parser · metrics.py Metric Scoring · rgb RGB Images
Components 11
Evaluation Inputs
- RGB Images
data/rgb - Thermal Images
data/infrared - Point Annotations
data/GT - YOLO Weights
Evaluation Pipeline
- Baseline Runner
src/baseline.py - GT Parser
src/gt_parser.py - Person Detector
src/detect.py - Point Matcher
src/matcher.py
Metrics and Results
- Metric Scoring
src/metrics.py - Run Reports
Other components
- Researcher
Connections 11
- Researcher → Baseline Runner: runs
README.md - Baseline Runner → GT Parser: loads annotations
src/baseline.py - GT Parser → Point Annotations: parses XML
src/gt_parser.py - Baseline Runner → RGB Images: reads frames
src/baseline.py - Baseline Runner → Thermal Images: reads frames
src/baseline.py - Baseline Runner → YOLO Weights: loads model
src/baseline.py - Baseline Runner → Person Detector: requests detections
src/baseline.py - Baseline Runner → Point Matcher: matches points
src/baseline.py - Baseline Runner → Metric Scoring: scores frames
src/baseline.py - Baseline Runner → Metric Scoring: aggregates scores
src/baseline.py - Baseline Runner → Run Reports: writes reports
src/baseline.py