File size: 6,435 Bytes
e8f11cf
df36123
 
 
 
 
 
 
 
 
 
e8f11cf
 
 
 
df36123
 
 
af6a58f
017a8ad
e8f11cf
 
 
 
 
 
 
 
af6a58f
 
e8f11cf
 
 
 
 
 
 
 
 
af6a58f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e8f11cf
af6a58f
 
 
 
 
 
 
 
 
 
 
 
 
 
e8f11cf
af6a58f
 
 
 
 
 
 
 
 
 
 
e8f11cf
af6a58f
 
e8f11cf
af6a58f
 
 
 
 
 
 
e8f11cf
af6a58f
 
 
 
 
 
e8f11cf
af6a58f
e8f11cf
af6a58f
 
 
 
 
 
 
 
 
 
e8f11cf
af6a58f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e8f11cf
af6a58f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e8f11cf
 
af6a58f
 
e8f11cf
af6a58f
017a8ad
e8f11cf
2b67239
e8f11cf
2b67239
 
e8f11cf
 
af6a58f
017a8ad
e8f11cf
 
 
 
 
 
 
 
 
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
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
---
license: mit
task_categories:
  - image-classification
language:
  - en
tags:
  - wildlife
  - camera-trap
  - animals
  - vision
  - zero-shot
  - clustering
  - benchmark
  - HUGO-Bench
size_categories:
  - 100K<n<1M
---

# HUGO-Bench

**Hierarchical Unsupervised Grouping of Organisms Benchmark**

A comprehensive benchmark dataset for evaluating zero-shot clustering of wildlife camera trap images using Vision Transformer embeddings.

## Overview

HUGO-Bench contains **139,111 expert-validated cropped images** of 60 animal species (30 birds, 30 mammals), derived from 23 camera trap projects across [LILA BC](https://lila.science/). The dataset enables benchmarking of Vision Transformer models for unsupervised species-level clustering without requiring labeled training data.

## Dataset Details

| Property | Value |
|----------|-------|
| **Format** | Parquet (optimized for streaming) |
| **Total Images** | 139,111 |
| **Species Count** | 60 species (30 birds, 30 mammals) |
| **Validated** | 138,024 (99.2%) |
| **Uncertain** | 1,087 (0.8%) |
| **License** | MIT |

### Splits
- `aves`: 73,528 images of 30 bird species
- `mammals`: 65,583 images of 30 mammal species

### Columns
- `image`: PIL Image object
- `label`: Taxonomic class (Aves or Mammals)
- `filename`: Original filename with species identifier (e.g., "american-crow_0001.jpg")

### Species Organization
Filenames follow the pattern: `{species-name}_{number}.jpg` or `uncertain_{species-name}_{number}.jpg`
- Species names use lowercase with hyphens (e.g., "american-crow", "red-fox")
- "uncertain_" prefix indicates cases where manual validation was uncertain of the given class

## Usage

### Streaming (recommended for large datasets)
```python
from datasets import load_dataset

# Load with streaming enabled
dataset = load_dataset("AI-EcoNet/HUGO-Bench", split="aves", streaming=True)

# Iterate through samples
for sample in dataset:
    image = sample['image']
    label = sample['label']
    filename = sample['filename']
    # Process your data
```

### Filter by species (streaming)
```python
from datasets import load_dataset

# Stream and filter for specific species
ds = load_dataset("AI-EcoNet/HUGO-Bench", split="aves", streaming=True)
crows_only = ds.filter(lambda x: x['filename'].startswith('american-crow'))

for sample in crows_only:
    print(sample['filename'])
```

### Load specific subset
```python
from datasets import load_dataset

# Load only first 1000 images from aves
dataset = load_dataset("AI-EcoNet/HUGO-Bench", split="aves[:1000]")

# Load 10% of mammals
dataset = load_dataset("AI-EcoNet/HUGO-Bench", split="mammals[:10%]")
```

### Full download
```python
from datasets import load_dataset

# Download full dataset
dataset = load_dataset("AI-EcoNet/HUGO-Bench")

# Access by split
aves_data = dataset['aves']
mammals_data = dataset['mammals']
```

## Complete Species List

### Aves (Birds) - 30 Species, 73,528 Images

| Species | Total | Validated | Uncertain |
|---------|-------|-----------|-----------|
| American crow | 1,240 | 1,234 | 6 |
| Australasian swamphen | 2,772 | 2,772 | 0 |
| Australian magpie | 2,739 | 2,733 | 6 |
| Black curassow | 1,949 | 1,939 | 10 |
| Blue whistling thrush | 2,157 | 2,128 | 29 |
| Brown quail | 2,662 | 2,652 | 10 |
| Chicken | 1,647 | 1,629 | 18 |
| Common chaffinch | 2,384 | 2,370 | 14 |
| Common myna | 2,707 | 2,704 | 3 |
| Dunnock | 2,066 | 2,055 | 11 |
| European starling | 1,876 | 1,867 | 9 |
| Fantails | 2,127 | 2,121 | 6 |
| Greenfinch | 2,503 | 2,483 | 20 |
| Kea | 2,833 | 2,814 | 19 |
| Kiwi | 1,634 | 1,620 | 14 |
| Kori bustard | 1,485 | 1,484 | 1 |
| Mountain quail | 2,381 | 2,337 | 44 |
| New Zealand robin | 2,753 | 2,719 | 34 |
| Ostrich | 3,727 | 3,722 | 5 |
| Petrel | 2,370 | 2,354 | 16 |
| Pipit | 2,316 | 2,238 | 78 |
| Red junglefowl | 2,684 | 2,671 | 13 |
| Spix's guan | 3,032 | 3,022 | 10 |
| Swamp harrier | 2,917 | 2,917 | 0 |
| Takahē | 2,085 | 2,073 | 12 |
| Tūī | 2,767 | 2,747 | 20 |
| Vulturine guineafowl | 4,420 | 4,404 | 16 |
| Weka | 2,421 | 2,407 | 14 |
| Wild turkey | 2,057 | 2,057 | 0 |
| Yellow-eyed penguin | 2,817 | 2,815 | 2 |

### Mammals - 30 Species, 65,583 Images

| Species | Total | Validated | Uncertain |
|---------|-------|-----------|-----------|
| Alpaca | 1,864 | 1,864 | 0 |
| American black bear | 2,912 | 2,797 | 115 |
| Black-backed jackal | 2,067 | 2,053 | 14 |
| Black rhinoceros | 1,428 | 1,419 | 9 |
| Bushpig | 1,241 | 1,230 | 11 |
| Common brushtail possum | 1,365 | 1,358 | 7 |
| Crab-eating mongoose | 2,364 | 2,316 | 48 |
| Crested porcupine | 1,449 | 1,449 | 0 |
| Dromedary camel | 1,809 | 1,798 | 11 |
| Eastern gray squirrel | 2,434 | 2,414 | 20 |
| Ferret badger | 1,743 | 1,671 | 72 |
| Gemsbok | 3,963 | 3,944 | 19 |
| Giant armadillo | 463 | 461 | 2 |
| Giraffe | 2,985 | 2,974 | 11 |
| Greater kudu | 4,410 | 4,375 | 35 |
| Hippopotamus | 2,319 | 2,314 | 5 |
| Jaguar | 6,855 | 6,831 | 24 |
| L'Hoest's monkey | 2,941 | 2,921 | 20 |
| Least weasel | 1,869 | 1,869 | 0 |
| Northern treeshrew | 699 | 687 | 12 |
| NZ sea lion | 2,296 | 2,277 | 19 |
| Raccoon | 2,700 | 2,577 | 123 |
| Serval | 625 | 625 | 0 |
| Ship rat | 822 | 822 | 0 |
| Spotted paca | 1,715 | 1,673 | 42 |
| Stump-tailed macaque | 3,558 | 3,535 | 23 |
| Sun bear | 713 | 709 | 4 |
| Warthog | 1,671 | 1,666 | 5 |
| White-nosed coati | 1,996 | 1,982 | 14 |
| Wolf | 2,307 | 2,307 | 0 |

**Note**: "Uncertain" indicates cases where manual validation was uncertain of the initial label.

## Citation

If you use HUGO-Bench, please cite:

```bibtex
@dataset{hugo_bench_2026,
  author       = {Markoff, Hugo and Bengtson, Stefan Hein and Ørsted, Michael},
  title        = {HUGO-Bench: Hierarchical Unsupervised Grouping of Organisms Benchmark},
  year         = {2026},
  publisher    = {Hugging Face Datasets},
  url          = {https://huggingface.co/datasets/AI-EcoNet/HUGO-Bench},
  note         = {Expert-validated camera-trap images for zero-shot clustering. Image sources from LILA BC.}
}
```

## Links

-  [Interactive Demo](https://hugomarkoff.github.io/animal_visual_transformer/)
-  Paper: *Vision Transformers for Zero-Shot Clustering of Animal Images: A Comparative Benchmarking Study*

## Acknowledgments

Raw image data derived from 23 projects on [LILA BC](https://lila.science/) (Labeled Information Library of Alexandria: Biology and Conservation). Please cite the original datasets as appropriate.