BirdNET Geomodel v3.0.4
Geographic species occurrence model from the BirdNET project. Predicts which species are likely at a given location and time of year.
This repository hosts the ONNX builds for use by BirdNET-Go as a range filter for its acoustic classifiers. Upstream patch releases are hosted side by side: the current release is v3.0.4, and the v3.0.2 files stay available for existing installs. models.json lists every file with its release, checksums and successor (see Releases below).
Model Details (v3.0.4)
- Version: V3.0.4
- Species: 14,082
- Input:
[latitude, longitude, week]as float32 (shape[batch, 3])- Latitude: [-90, 90]
- Longitude: [-180, 180]
- Week: BirdNET 48-week year (1-48, 4 weeks per month)
- Output:
probabilities, 14,082 float32 species occurrence scores (sigmoid) - Format: ONNX (opset 18), FP32 and FP16
Files
v3.0.4 (current)
| File | Description | Size |
|---|---|---|
birdnet-geomodel-v3.0.4-fp32-b1.onnx |
ONNX FP32. Recommended: more accurate, and lower CPU RAM than FP16 (ONNX Runtime upconverts FP16 on CPU). | 15.5 MB |
birdnet-geomodel-v3.0.4-fp16-b1.onnx |
ONNX FP16 | 7.9 MB |
birdnet-geomodel-v3.0.4-labels-b1.txt |
Species labels, <scientific name>_<common name> per line (14,082 entries) |
565 KB |
birdnet-geomodel-v3.0.4-taxonomy-b1.csv |
Taxonomy with multilingual common names (16,171 entries, covers every v3.0.4 class) | 11.1 MB |
BirdNET+_Geomodel_V3.0.4_Global_14K_Labels.txt |
Upstream labels file as released | 672 KB |
v3.0.2 (kept for existing installs)
| File | Description | Size |
|---|---|---|
birdnet-geomodel-v3.0-fp32.onnx |
ONNX FP32 (holds V3.0.2 weights; frozen filename) | 14.7 MB |
birdnet-geomodel-v3.0-fp16.onnx |
ONNX FP16 (holds V3.0.2 weights; frozen filename) | 7.5 MB |
BirdNET+_Geomodel_V3.0.2_Global_12K_FP16.onnx |
Same bytes as birdnet-geomodel-v3.0-fp16.onnx, upstream filename |
7.5 MB |
geomodel_v3.0.2_labels.txt |
Species labels, <scientific name>_<common name> per line (12,012 entries) |
479 KB |
BirdNET+_Geomodel_V3.0.2_Global_12K_Labels.txt |
Upstream labels file as released | 572 KB |
taxonomy.csv |
Taxonomy with multilingual common names (13,361 entries) | 9.2 MB |
Shared
| File | Description |
|---|---|
models.json |
Machine-readable index: every file with release, variant, checksums, backends and successor |
LICENSE-MODELS.md |
Apache License 2.0 for the model artifacts |
ACCEPTABLE_USE.md |
BirdNET Acceptable Use Policy (guidance, not a license condition) |
SHA256SUMS |
File integrity checksums |
Labels format
The <scientific name>_<common name> labels files have one line per model output (line 1 = index 0). The upstream BirdNET+_..._Labels.txt files are tab-separated species_code, scientific_name, common_name in the same order.
Taxonomy CSV
Header row, then one row per species: index, scientific name, English common name, species code, class, and common names in other languages (common_name_<locale> columns). Read columns by header name: the column set and order differ between the v3.0.2 and v3.0.4 files.
Releases and updates
The ONNX files are the upstream release files, renamed to birdnet-geomodel-v<version>-<variant>-b<build>.onnx. A published file is never changed or removed. models.json carries the release index in family_meta.versions and the current release in family_meta.current_version; each older file's superseded_by names the same-variant file of the next release, so a client can follow the chain from an installed file to the current one.
Source
Original models from birdnet-team/geomodel v3.0.4 and v3.0.2. The v3.0.4 taxonomy is taxonomy_v0.2-Jun2026.csv from the upstream repository at the v3.0.4 tag.
License
Apache License 2.0. The BirdNET team licenses the Geomodel model artifacts (checkpoints, converted models, and labels) under Apache 2.0; see LICENSE-MODELS.md.
The BirdNET Acceptable Use Policy describes expectations for responsible use. It is guidance and does not add conditions to the Apache 2.0 license.
Attribution: Powered by BirdNET.
Citation
If you use this model in publications or presentations, please cite:
Powered by BirdNET (https://birdnet.cornell.edu/)