BGLab commited on
Commit
91b2ee7
Β·
verified Β·
1 Parent(s): ccbb87b

added readme

Browse files
Files changed (1) hide show
  1. README.md +51 -53
README.md CHANGED
@@ -2,65 +2,63 @@
2
  License: cc0-1.0
3
  language:
4
  - en
5
- pretty_name: Arboretum
6
  task_categories:
7
- - image-classification
8
- - zero-shot-classification
9
  tags:
10
- - biology
11
- - image
12
- - animals
13
- - species
14
- - taxonomy
15
- - rare species
16
- - endangered species
17
- - evolutionary biology
18
- - balanced
19
- - CV
20
- - multimodal
21
- - CLIP
22
- - knowledge-guided
23
-
24
  size_categories: 100M<n<1B
25
-
26
  ---
27
 
28
- # Arboretum: A Comprehensive Multimodal Dataset Enabling AI for Biodiversity
29
 
30
  <!-- Banner links -->
31
  <div style="text-align:center;">
32
- <a href="https://baskargroup.github.io/Arboretum/" target="_blank">
33
  <img src="https://img.shields.io/badge/Project%20Page-Visit-blue" alt="Project Page" style="margin-right:10px;">
34
  </a>
35
- <a href="https://github.com/baskargroup/Arboretum" target="_blank">
36
  <img src="https://img.shields.io/badge/GitHub-Visit-lightgrey" alt="GitHub" style="margin-right:10px;">
37
  </a>
38
  <a href="https://pypi.org/project/arbor-process/" target="_blank">
39
- <img src="https://img.shields.io/badge/PyPI-arbor--process%200.1.0-orange" alt="PyPI arbor-process 0.1.0">
40
  </a>
41
  </div>
42
 
43
  ## Description
44
 
45
- [Arboretum](https://baskargroup.github.io/Arboretum/) comprises well-processed metadata with full taxa information and URLs pointing to image files. The metadata can be used to filter specific categories, visualize data distribution, and manage imbalance effectively. We provide a collection of software tools that enable users to easily download, access, and manipulate the dataset.
46
 
47
 
48
- ## Arboretum Dataset
49
- `Arboretum` comprises over `134.6M` images across seven taxonomic classes β€”Aves, Arachnida, Insecta, Plantae, Fungi, Mollusca, and Reptilia.
50
- These taxonomic classes were chosen to represent the span of species β€” outside of charismatic megafauna. The images in Arboretum span `326,888`species.
51
- Overall, this dataset nearly matches the state-of-the-art curated dataset (TREEOFLIFE-10M) in terms of species diversity, while comfortably exceeding it in terms of scale by a factor of nearly 13.5 times.
52
 
53
  ## New Benchmark Datasets
54
  We created three new benchmark datasets for fine-grained image classification. In addition, we provide a new benchmark dataset for species recognition across various developmental Life-stages.
55
 
56
- ### Arboretum-Balanced
57
- For balanced species distribution across the 7 categories, we curated `Arboretum-Balanced`. Each category includes up to 500 species, with 50 images per species.
58
 
59
- ### Arboretum-Unseen
60
- To provide a robust benchmark for evaluating the generalization capability of models on unseen species, we curated `Arboretum-Unseen`. The test dataset was constructed by identifying species with fewer than 30 instances in ARBORETUM, ensuring that the dataset contains species that were unseen by ARBORCLIP. Each species contained 10 images.
61
 
62
- ### Arboretum-LifeStages
63
- To assess the model’s ability to recognize species across various developmental stages, we curated `Arboretum-LifeStages`. This dataset has 20 labels in total and focuses on insects, since these species often exhibit significant visual differences across their lifespan. Arboretum-LifeStages contains five insect species and utilized the observation export feature on the iNaturalist platform to collect data from 2/1/2024 to 5/20/2024 to ensure no overlap with the training dataset. For each species, life stage filters (egg, larva, pupa, or adult) were applied.
64
 
65
  ## Dataset Information
66
 
@@ -72,42 +70,42 @@ To assess the model’s ability to recognize species across various developmenta
72
  - **Life Stages Dataset**: Focuses on insects across various developmental stages.
73
 
74
 
75
- ## ArborCLIP Models
76
 
77
- **See the [ArborCLIP](https://huggingface.co/ChihHsuan-Yang/ArborCLIP) model card on HuggingFace to download the trained model checkpoints**
78
 
79
- We released three trained model checkpoints in the [ArborCLIP](https://huggingface.co/ChihHsuan-Yang/ArborCLIP) model card on HuggingFace. These CLIP-style models were trained on [ARBORETUM-40M](https://baskargroup.github.io/Arboretum/) for the following configurations:
80
 
81
- - **ARBORCLIP-O:** Trained a ViT-B/16 backbone initialized from the [OpenCLIP's](https://github.com/mlfoundations/open_clip) checkpoint. The training was conducted for 40 epochs.
82
- - **ARBORCLIP-B:** Trained a ViT-B/16 backbone initialized from the [BioCLIP's](https://github.com/Imageomics/BioCLIP) checkpoint. The training was conducted for 8 epochs.
83
- - **ARBORCLIP-M:** Trained a ViT-L/14 backbone initialized from the [MetaCLIP's](https://github.com/facebookresearch/MetaCLIP) checkpoint. The training was conducted for 12 epochs.
84
 
85
 
86
 
87
  ## Usage
88
 
89
- **To start using the Arboretum dataset, follow the instructions provided in the [GitHub](https://github.com/baskargroup/Arboretum). Model checkpoints are shared in the [model_ckpt](#directory) directory.**
90
 
91
- **Metadata files are included in the [Directory](#directory). Please download the metadata from the [Directory](#directory)** and pre-process the data using the [arbor_process](https://pypi.org/project/arbor-process/) PyPI library. The instructions to use the library can be found in [here](https://github.com/baskargroup/Arboretum/blob/main/Arbor-preprocess/README_arbor_process.md). The Readme file contains the detailed description of data preparation steps.
92
 
93
  ### Directory
94
  ```plaintext
95
  main/
96
- β”œβ”€β”€ Arboretum/
97
  β”‚ β”œβ”€β”€ chunk_0.csv
98
  β”‚ β”œβ”€β”€ chunk_0.parquet
99
  β”‚ β”œβ”€β”€ chunk_1.parquet
100
  β”‚ β”œβ”€β”€ .
101
  β”‚ β”œβ”€β”€ .
102
  β”‚ β”œβ”€β”€ .
103
- β”‚ └── chunk_2691.parquet
104
- β”œβ”€β”€ Arboretum-benchmark/
105
- β”‚ β”œβ”€β”€ Arboretum-Balanced.csv
106
- β”‚ β”œβ”€β”€ Arboretum-Balanced.parquet
107
- β”‚ β”œβ”€β”€ Arboretum-Lifestages.csv
108
- β”‚ β”œβ”€β”€ Arboretum-Lifestages.parquet
109
- β”‚ β”œβ”€β”€ Arboretum-Unseen.csv
110
- β”‚ └──Arboretum-Unseen.parquet
111
  β”œβ”€β”€README.md
112
  └──.gitignore
113
  ```
@@ -124,8 +122,8 @@ expertise.
124
  <div class="container is-max-widescreen content">
125
  <h2 class="title">Citation</h2>
126
  If you find this dataset useful in your research, please consider citing our paper:
127
- <pre><code>@misc{yang2024arboretumlargemultimodaldataset,
128
- title={Arboretum: A Large Multimodal Dataset Enabling AI for Biodiversity},
129
  author={Chih-Hsuan Yang, Benjamin Feuer, Zaki Jubery, Zi K. Deng, Andre Nakkab,
130
  Md Zahid Hasan, Shivani Chiranjeevi, Kelly Marshall, Nirmal Baishnab, Asheesh K Singh,
131
  Arti Singh, Soumik Sarkar, Nirav Merchant, Chinmay Hegde, Baskar Ganapathysubramanian},
@@ -141,4 +139,4 @@ expertise.
141
 
142
  ---
143
 
144
- For more details and access to the dataset, please visit the [Project Page](https://baskargroup.github.io/Arboretum/).
 
2
  License: cc0-1.0
3
  language:
4
  - en
5
+ pretty_name: BioTrove
6
  task_categories:
7
+ - image-classification
8
+ - zero-shot-classification
9
  tags:
10
+ - biology
11
+ - image
12
+ - animals
13
+ - species
14
+ - taxonomy
15
+ - rare species
16
+ - endangered species
17
+ - evolutionary biology
18
+ - balanced
19
+ - CV
20
+ - multimodal
21
+ - CLIP
22
+ - knowledge-guided
 
23
  size_categories: 100M<n<1B
 
24
  ---
25
 
26
+ # BioTrove: A Large Curated Image Dataset Enabling AI for Biodiversity
27
 
28
  <!-- Banner links -->
29
  <div style="text-align:center;">
30
+ <a href="https://baskargroup.github.io/BioTrove/ target="_blank">
31
  <img src="https://img.shields.io/badge/Project%20Page-Visit-blue" alt="Project Page" style="margin-right:10px;">
32
  </a>
33
+ <a href="https://github.com/baskargroup/BioTrove" target="_blank">
34
  <img src="https://img.shields.io/badge/GitHub-Visit-lightgrey" alt="GitHub" style="margin-right:10px;">
35
  </a>
36
  <a href="https://pypi.org/project/arbor-process/" target="_blank">
37
+ <img src="https://img.shields.io/badge/PyPI-arbor--process%200.1.0-orange" alt="PyPI biotrove-process 0.1.0">
38
  </a>
39
  </div>
40
 
41
  ## Description
42
 
43
+ [BioTrove](https://baskargroup.github.io/BioTrove/) comprises well-processed metadata with full taxa information and URLs pointing to image files. The metadata can be used to filter specific categories, visualize data distribution, and manage imbalance effectively. We provide a collection of software tools that enable users to easily download, access, and manipulate the dataset.
44
 
45
 
46
+ ## BioTrove Dataset
47
+ `BioTrove` comprises over `161.9M` images across several taxonomic groups- including Reptilia (reptiles), Plantae (plants), Mollusca (mollusks), Mammalia (mammals), Insecta (insects), Fungi (fungi), Aves (birds), Arachnida (arachnids), Animalia (animals), Amphibia (amphibians), and Actinopterygii (ray-finned fish).
48
+ These taxonomic groups were chosen to represent the span of species β€” outside of charismatic megafauna. The images in BioTrove span `366.6K`species.
49
+ Overall, this dataset nearly matches the state-of-the-art curated dataset (TREEOFLIFE-10M) in terms of species diversity, while comfortably exceeding it in terms of scale by a factor of nearly 16.2 times.
50
 
51
  ## New Benchmark Datasets
52
  We created three new benchmark datasets for fine-grained image classification. In addition, we provide a new benchmark dataset for species recognition across various developmental Life-stages.
53
 
54
+ ### BioTrove-Balanced
55
+ For balanced species distribution across the 7 categories, we curated `BioTrove-Balanced`. Each category includes up to 500 species, with 50 images per species.
56
 
57
+ ### BioTrove-Unseen
58
+ To provide a robust benchmark for evaluating the generalization capability of models on unseen species, we curated `BioTrove-Unseen`. The test dataset was constructed by identifying species with fewer than 30 instances in BioTrove, ensuring that the dataset contains species that were unseen by BioTrove-CLIP. Each species contained 10 images.
59
 
60
+ ### BioTrove-LifeStages
61
+ To assess the model’s ability to recognize species across various developmental stages, we curated `BioTrove-LifeStages`. This dataset has 20 labels in total and focuses on insects, since these species often exhibit significant visual differences across their lifespan. BioTrove-LifeStages contains five insect species and utilized the observation export feature on the iNaturalist platform to collect data from 2/1/2024 to 5/20/2024 to ensure no overlap with the training dataset. For each species, life stage filters (egg, larva, pupa, or adult) were applied.
62
 
63
  ## Dataset Information
64
 
 
70
  - **Life Stages Dataset**: Focuses on insects across various developmental stages.
71
 
72
 
73
+ ## BioTrove-CLIP Models
74
 
75
+ **See the [BioTrove-CLIP](https://huggingface.co/BGLab/BioTrove-CLIP) model card on HuggingFace to download the trained model checkpoints**
76
 
77
+ We released three trained model checkpoints in the [BioTrove-CLIP](https://huggingface.co/BGLab/BioTrove-CLIP) model card on HuggingFace. These CLIP-style models were trained on [BioTrove-40M](https://baskargroup.github.io/BioTrove/) for the following configurations:
78
 
79
+ - **BioTrove-CLIP-O:** Trained a ViT-B/16 backbone initialized from the [OpenCLIP's](https://github.com/mlfoundations/open_clip) checkpoint. The training was conducted for 40 epochs.
80
+ - **BioTrove-CLIP-B:** Trained a ViT-B/16 backbone initialized from the [BioCLIP's](https://github.com/Imageomics/BioCLIP) checkpoint. The training was conducted for 8 epochs.
81
+ - **BioTrove-CLIP-M:** Trained a ViT-L/14 backbone initialized from the [MetaCLIP's](https://github.com/facebookresearch/MetaCLIP) checkpoint. The training was conducted for 12 epochs.
82
 
83
 
84
 
85
  ## Usage
86
 
87
+ **To start using the BioTrove dataset, follow the instructions provided in the [GitHub](https://github.com/baskargroup/BioTrove). Model checkpoints are shared in the [model_ckpt](#directory) directory.**
88
 
89
+ **Metadata files are included in the [Directory](#directory). Please download the metadata from the [Directory](#directory)** and pre-process the data using the [arbor_process](https://pypi.org/project/arbor-process/) PyPI library. The instructions to use the library can be found in [here](https://github.com/baskargroup/BioTrove/blob/main/Arbor-preprocess/README_arbor_process.md). The Readme file contains the detailed description of data preparation steps.
90
 
91
  ### Directory
92
  ```plaintext
93
  main/
94
+ β”œβ”€β”€ BioTrove/
95
  β”‚ β”œβ”€β”€ chunk_0.csv
96
  β”‚ β”œβ”€β”€ chunk_0.parquet
97
  β”‚ β”œβ”€β”€ chunk_1.parquet
98
  β”‚ β”œβ”€β”€ .
99
  β”‚ β”œβ”€β”€ .
100
  β”‚ β”œβ”€β”€ .
101
+ β”‚ └── chunk_3251.parquet
102
+ β”œβ”€β”€ BioTrove-benchmark/
103
+ β”‚ β”œβ”€β”€ BioTrove-Balanced.csv
104
+ β”‚ β”œβ”€β”€ BioTrove-Balanced.parquet
105
+ β”‚ β”œβ”€β”€ BioTrove-Lifestages.csv
106
+ β”‚ β”œβ”€β”€ BioTrove-Lifestages.parquet
107
+ β”‚ β”œβ”€β”€ BioTrove-Unseen.csv
108
+ β”‚ └──BioTrove-Unseen.parquet
109
  β”œβ”€β”€README.md
110
  └──.gitignore
111
  ```
 
122
  <div class="container is-max-widescreen content">
123
  <h2 class="title">Citation</h2>
124
  If you find this dataset useful in your research, please consider citing our paper:
125
+ <pre><code>@misc{yang2024BioTrovelargemultimodaldataset,
126
+ title={BioTrove: A Large Multimodal Dataset Enabling AI for Biodiversity},
127
  author={Chih-Hsuan Yang, Benjamin Feuer, Zaki Jubery, Zi K. Deng, Andre Nakkab,
128
  Md Zahid Hasan, Shivani Chiranjeevi, Kelly Marshall, Nirmal Baishnab, Asheesh K Singh,
129
  Arti Singh, Soumik Sarkar, Nirav Merchant, Chinmay Hegde, Baskar Ganapathysubramanian},
 
139
 
140
  ---
141
 
142
+ For more details and access to the dataset, please visit the [Project Page](https://baskargroup.github.io/BioTrove/).