OrcaTCA β€” results and model weights

Killer whale acoustic detection and transfer analysis using the DCLDE archive. Error rates refer to annotated five-second windows; false-positive rate is not the number of alarms per hour.

Contents

  • result/: 107 files (figures, metrics and analyses).
  • runs/: 705 files (training runs and checkpoints).
  • state/: 80 files (excluding cache/ directories).
  • 18 best.pt checkpoints, including any found outside those directories.
  • 1.13 GiB of original published artifacts.
  • model_index.csv: weight locations, fold/seed and SHA256.
  • release_manifest.json: complete file checksums.

Model weights

Weights are original project-specific PyTorch checkpoints. Per-run metrics are stored alongside their runs when present. A single checkpoint does not represent the study-wide mean across runs or seeds.

Location Weights
semantic230_20261029T160625Z_9e1b11a8 18

Results

The saved 1% source-FPR transfer analysis and its published figure inputs were checked against their recorded SHA256 hashes.

Main results

result/analysis/transfer_review/main_results.csv

method macro_pauc
natural_coarse_mixstyle 0.7919675938206293
natural_mixstyle 0.779366374915182
natural_subtype_mixstyle 0.8098009556475368
natural_threeclass_mixstyle 0.7995478111199827

Transfer gap

result/analysis/transfer_review/transfer_gap_summary.csv

target_units target_boundary_labels target_negative_windows target_recall max_abs_target_block_gap_change target_block_removals_valid target_block_removals_not_estimable
62 152 0.7081659445976107 0.09608733745953876 19 0
62 152 0.6824340291453327 0.08174186585756543 19 0
62 152 0.7448995667585666 0.12449611420136115 19 0
62 152 0.7371668635945909 0.1031161812281778 19 0
62 152 0.8732309308126559 0.1678662754782957 19 0

Paired pAUC

result/figure_data/derived/pauc_gain.csv

site_selection fine_minus_binary_pauc pauc_difference_times_100
all_sites 0.030434580732354832 3.0434580732354832
at_least_20 0.004699497448187984 0.46994974481879837

The tables are previews only. Source recordings are not included. Release paths preserve the original project layout.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support