Bird Counting


Results

Task description

This task is part of the BioDCASE Challenge 2026 and focuses on estimating bird abundance from acoustic recordings collected in zoo aviaries. The goal is to develop robust methods for counting individuals of a target species in realistic, multi-species acoustic environments where ground-truth population sizes are known. The main leaderboard ranks systems based solely on target-species abundance estimation performance.

Task description page

Awards

The official winning submission is determined by the official evaluation ranking. The Jury Award is selected by the task organisers based on the submitted systems and technical reports.

Official winning submission

Yang, Jing and Li, Siyi and Zheng, Zihan and Zhang, Heng and Ni, Youran and Wang, Yuzhu and Huang, Gongping
Wuhan University; Beijing Jiaotong University; Tampere University
A Two-Stage CRNN Regression Framework for Bird Counting

Jury award

Leon Rojas, Yuneva E. and Frery, Alejandro C. and Colonna, Juan G.
Universidade Federal do Amazonas; Victoria University of Wellington
Acoustic Bird Abundance Estimation Using Morphological Features of the Spectrogram
Awarded for the band-filtered stochastic ratio, an original and detector-independent morphological feature that directly measures flamingo acoustic saturation (r = 0.9978), delivered as a fully open-source, reproducible pipeline that also placed 2nd.

Submission ranking

The official submission ranking is based on the average MAE across each of the test sets. Individual performance on each test set is reported below.

Submission information Rank Test set Dev set
Submission System name Technical
report
Official
rank
Rank
score
MAE Flamingo MAE Quelea MAE Ibis RMSE R2 MAPE Avg MAE
(dev)
Yang_WHU_task6_1 Two-Stage CRNN Regression Framework yangCRNN2026 1 9.17 20.00 4.50 3.00 13.04 0.97 71.68 0.75
Leon_UFAM_task6_3 ARIA fCWR x Stochastic Band Ratio (linear) leonMorphology2026 2 13.33 23.50 6.00 10.50 18.02 0.94 103.98 2.25
You_PKU_task6_2 Low-Parameter Detector Ensemble (stacked flamingo) youDetector2026 3 22.67 61.50 3.50 3.00 48.69 0.59 207.23 0.00
official_aria_baseline Official ARIA Baseline (flock_corrected_cwr) arginBaseline2026 4 24.50 69.00 1.50 3.00 51.35 0.55 56.20 11.50
Zhong_NJU_task6_1 Validation-Gated Sampling-Based ARIA Counting zhongARIA2026 5 46.67 83.00 40.50 16.50 58.99 0.40 234.95 11.50

Technical reports

Acoustic Bird Abundance Estimation Using Morphological Features of the Spectrogram

Leon Rojas, Yuneva E. and Frery, Alejandro C. and Colonna, Juan G.
Universidade Federal do Amazonas; Victoria University of Wellington

Method summary
Features flock-corrected CWR; band-filtered stochastic ratio (mathematical morphology)
Decision rule least-squares linear regression on fCWR x stochastic band ratio
Code
Source code

A Two-Stage CRNN Regression Framework for Bird Counting

Yang, Jing and Li, Siyi and Zheng, Zihan and Zhang, Heng and Ni, Youran and Wang, Yuzhu and Huang, Gongping
Wuhan University; Beijing Jiaotong University; Tampere University

Method summary
Features log-mel spectrogram (128-band)
Training or augmentation SpecAugment; synthetic data mixing
Decision rule fragment-level CRNN regression + aviary-level linear aggregation

Low-Parameter Detector Ensembles for Bird Counting

You, Yuhuan
Peking University

Method summary
Features detection rate; confidence-weighted rate; bout statistics; MAAD acoustic indices
Embeddings BirdNET; ARIA
Decision rule species-specific low-parameter regression

Validation-Gated Sampling-Based ARIA Counting System

Zhong, Chongyang
Nanjing University

Method summary
Features confidence-weighted detection rate
Decision rule stratified sampling + linear calibration + species-wise validation gating