File size: 9,451 Bytes
9d87076 0ad5785 9d87076 0ad5785 9d87076 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 | ---
pretty_name: BirdNET+ V3.0 (GPU-native full + regional catalog)
license: apache-2.0
tags:
- bioacoustics
- birdnet
- bird
- vocalization
- audio-classification
- onnx
library_name: onnx
---
# BirdNET+ V3.0 (GPU-native full + regional catalog)
BirdNET+ V3.0 acoustic classifier, re-exported to a
**GPU-native ONNX** graph, plus a catalog of **region-specific slices** that are
much smaller while staying numerically identical to the full model on the
species they keep.
## Licence and terms
Each release came with its own upstream terms, listed here with its licence. The
repo-level licence tag is the current release's.
- **BirdNET+ V3.0** (`birdnet-plus-v3.0`), Apache-2.0: the Apache-2.0 model licence and acceptable-use guidance of the BirdNET+ V3.0 release (Zenodo, https://doi.org/10.5281/zenodo.24162518), `LICENSE-v3.0.md` and `ACCEPTABLE_USE-v3.0.md` at the repo root, unchanged from upstream.
## Origin and attribution
**Powered by BirdNET** (https://birdnet.cornell.edu/). BirdNET+ V3.0 is developed
by the **K. Lisa Yang Center for Conservation Bioacoustics (Cornell University)**,
**Chemnitz University of Technology**, and the **Museum fuer Naturkunde Berlin**.
It builds on Kahl et al. (2021) and Lasseck (2018). Architecture of this release:
EfficientNetV2-S backbone, three-head classifier, 11,725 classes.
The GPU-native conversion and the regional slicing in this repo are derived work
under Apache-2.0.
## Releases in this repo
The current release is **BirdNET+ V3.0** (`birdnet-plus-v3.0`).
## What "GPU-native" means
The BirdNET team's ONNX export computes its log-mel front-end with an in-graph
ONNX `STFT` op. ONNX Runtime has no CUDA/TensorRT kernel for `STFT`, and OpenVINO
cannot translate it at all, so the stock export cannot run on GPU or OpenVINO. In
these models the `STFT` is replaced with a mathematically equivalent `Conv1d`
(framing + windowed DFT), so the graph holds no op those runtimes lack.
Every file was run on **ONNX Runtime CPU** (x86-64 and a Raspberry Pi 5) and on **OpenVINO** at f32 (CPU on both, and an Intel integrated GPU), and its scores were compared with the x86-64 ONNX Runtime scores of the same file; the manifest's `supported` and `verified_on` follow those runs. OpenVINO must run at f32: its default f16 hint gives wrong scores on the Cortex-A76 and on Intel GPUs. It is not recommended: its margin over ONNX Runtime CPU was 7% at most where it was measured. The CUDA and TensorRT entries were not run for this release. One model serves CPU and GPU; there is no separate CPU/GPU
build.
Input: raw audio `[batch, 160000]` (**5 s @ 32 kHz**, matching Perch v2 so a host
can feed both classifiers the same buffer). Outputs: `predictions [batch, 11,725]`
(a static dimension, sigmoid already applied) and `embeddings [batch, 1280]`.
## Variants and hardware
| precision | best for |
|---|---|
| `-fp32-` | **run this**: ONNX Runtime and OpenVINO (BirdNET-Go runs v3.0 at f32 on every OpenVINO device) |
This release ships FP32 only. For low-RAM CPU devices the right choice is a
**regional FP32** model.
int8-arm is not shipped for v3.0: the backbone is Conv-dominated and int8-arm only
quantizes MatMul/Gemm weights, so it barely helps here.
## Full model
| file | classes | MB |
|---|---:|---:|
| `full/birdnet-plus-v3.0-fp32-b1.onnx` | 11,725 | 361 |
| `full/birdnet-plus-v3.0-labels-b1.txt` | 11,725 | - |
## Why regional models
The three-head classifier holds a large share of the model's parameters, so
restricting it to a region's species set cuts memory and download size. Each tile
is **numerically identical** to the full model on the species it keeps (max abs
difference vs the full model < 1e-6); the only change is that out-of-region
species are not emitted, which at a fixed monitoring location is exactly what you
want. Region species sets come from the **BirdNET Geomodel v3.0.2** range filter
(species by occurrence probability, up to a per-tile cap, plus a cosmopolitan core),
using the same tiles as the Perch v2 regional catalog. The cap is 800 for most
temperate tiles and higher for species-rich ones. Every tile also contains Homo sapiens and the release's non-animal sound classes, whatever its range filter says.
A species the geomodel names differently from the classifier (a moved genus, a
changed spelling) is matched through a reviewed rename table, so the tiles keep
species an exact name match would have dropped.
## Regional catalog
Each tile ships an fp32 model + labels + indices + `coverage.png` +
metadata. `regional/groups.json` lists the continent grouping.
### Europe
| region | coverage | classes | fp32 MB |
|---|---|---:|---:|
| `azores` | <img src="regional/azores/coverage.png" width="210"> | 242 | 113.7 |
| `baltics` | <img src="regional/baltics/coverage.png" width="210"> | 460 | 118.4 |
| `british-isles` | <img src="regional/british-isles/coverage.png" width="210"> | 579 | 121.0 |
| `canary-islands` | <img src="regional/canary-islands/coverage.png" width="210"> | 412 | 117.4 |
| `central-europe` | <img src="regional/central-europe/coverage.png" width="210"> | 667 | 122.9 |
| `eastern-europe` | <img src="regional/eastern-europe/coverage.png" width="210"> | 540 | 120.1 |
| `iberia` | <img src="regional/iberia/coverage.png" width="210"> | 649 | 122.5 |
| `iceland` | <img src="regional/iceland/coverage.png" width="210"> | 409 | 117.3 |
| `madeira` | <img src="regional/madeira/coverage.png" width="210"> | 274 | 114.4 |
| `nordic` | <img src="regional/nordic/coverage.png" width="210"> | 443 | 118.0 |
| `southern-europe` | <img src="regional/southern-europe/coverage.png" width="210"> | 638 | 122.2 |
| `svalbard` | <img src="regional/svalbard/coverage.png" width="210"> | 298 | 114.9 |
| `western-palearctic` | <img src="regional/western-palearctic/coverage.png" width="210"> | 1418 | 139.0 |
### Asia
| region | coverage | classes | fp32 MB |
|---|---|---:|---:|
| `china-north-central` | <img src="regional/china-north-central/coverage.png" width="210"> | 791 | 125.5 |
| `china-northeast` | <img src="regional/china-northeast/coverage.png" width="210"> | 688 | 123.3 |
| `china-southeast` | <img src="regional/china-southeast/coverage.png" width="210"> | 1178 | 133.9 |
| `china-southwest` | <img src="regional/china-southwest/coverage.png" width="210"> | 1228 | 134.9 |
| `himalaya` | <img src="regional/himalaya/coverage.png" width="210"> | 1226 | 134.9 |
| `indo-gangetic` | <img src="regional/indo-gangetic/coverage.png" width="210"> | 1199 | 134.3 |
| `japan` | <img src="regional/japan/coverage.png" width="210"> | 609 | 121.6 |
| `south-asia-peninsular` | <img src="regional/south-asia-peninsular/coverage.png" width="210"> | 671 | 122.9 |
| `tibet` | <img src="regional/tibet/coverage.png" width="210"> | 1226 | 134.9 |
### North America
| region | coverage | classes | fp32 MB |
|---|---|---:|---:|
| `canada-alaska` | <img src="regional/canada-alaska/coverage.png" width="210"> | 777 | 125.2 |
| `north-america-east` | <img src="regional/north-america-east/coverage.png" width="210"> | 818 | 126.1 |
| `north-america-west` | <img src="regional/north-america-west/coverage.png" width="210"> | 821 | 126.2 |
### South America
| region | coverage | classes | fp32 MB |
|---|---|---:|---:|
| `amazonia` | <img src="regional/amazonia/coverage.png" width="210"> | 3182 | 177.0 |
| `andes` | <img src="regional/andes/coverage.png" width="210"> | 3314 | 179.8 |
| `eastern-brazil` | <img src="regional/eastern-brazil/coverage.png" width="210"> | 1972 | 150.9 |
| `galapagos` | <img src="regional/galapagos/coverage.png" width="210"> | 156 | 111.9 |
| `southern-cone` | <img src="regional/southern-cone/coverage.png" width="210"> | 1639 | 143.8 |
### Africa
| region | coverage | classes | fp32 MB |
|---|---|---:|---:|
| `cape-verde` | <img src="regional/cape-verde/coverage.png" width="210"> | 162 | 112.0 |
| `central-africa` | <img src="regional/central-africa/coverage.png" width="210"> | 1537 | 141.6 |
| `east-africa` | <img src="regional/east-africa/coverage.png" width="210"> | 1824 | 147.8 |
| `mauritius` | <img src="regional/mauritius/coverage.png" width="210"> | 90 | 110.4 |
| `reunion` | <img src="regional/reunion/coverage.png" width="210"> | 92 | 110.5 |
| `sao-tome-principe` | <img src="regional/sao-tome-principe/coverage.png" width="210"> | 133 | 111.4 |
| `seychelles` | <img src="regional/seychelles/coverage.png" width="210"> | 137 | 111.4 |
| `southern-africa` | <img src="regional/southern-africa/coverage.png" width="210"> | 820 | 126.1 |
| `west-africa` | <img src="regional/west-africa/coverage.png" width="210"> | 1027 | 130.6 |
### Oceania
| region | coverage | classes | fp32 MB |
|---|---|---:|---:|
| `australia-east` | <img src="regional/australia-east/coverage.png" width="210"> | 757 | 124.8 |
| `hawaii` | <img src="regional/hawaii/coverage.png" width="210"> | 294 | 114.8 |
| `new-caledonia` | <img src="regional/new-caledonia/coverage.png" width="210"> | 190 | 112.6 |
| `new-zealand` | <img src="regional/new-zealand/coverage.png" width="210"> | 289 | 114.7 |
## Deployment
- Point your host at a tile's model + its labels file (matching counts) and
restart. Labels are `Scientific name_Common name`.
- Confidence is a per-class sigmoid; slicing does not change a kept class's score
(unlike a softmax model), so thresholds transfer directly from the full model.
- Binaries are hosted here on HuggingFace; the derivation tooling lives separately.
|