Results¶
By default, BirdBox writes outputs under results/.
Recommended Layout¶
results/
raw_detections.json ← detect_birds --no-merge
with_algorithm_metadata.json ← filter_and_merge (optional)
simplified.csv ← filter_and_merge (confusion matrix input)
f_1.0_score_analysis/ ← f_beta_score_analysis outputs
confusion_matrix_analysis/ ← confusion matrix outputs
results/run_2/ ← created automatically if you re-run the pipeline
raw_detections.json
simplified.csv
...
When you run the default pipeline again while results/ already contains detection outputs, detect_birds.py writes to results/run_2/ (then run_3/, and so on) and stores the active path in results/.active_run. Later steps with default paths follow that marker.
Per-dataset runs (e.g. results/Hawaii/) use a single flat folder without auto-versioning.
Reproducibility Tips¶
Keep Your Results Reproducible
- Keep one folder per dataset/model configuration when comparing runs.
- Always preserve
raw_detections.json. It is the source for all threshold experiments and can be re-filtered without re-running inference. - Archive the exact
--confthreshold and--song-gapvalue used for any final exports.