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.