egrace479 commited on
Commit
051f42c
1 Parent(s): 8847695

Add information to Dataset Card.

Browse files
Files changed (1) hide show
  1. README.md +170 -40
README.md CHANGED
@@ -1,122 +1,252 @@
1
  ---
2
- task_categories:
3
- - zero-shot-classification
4
- - image-classification
5
  language:
6
  - en
 
 
 
 
 
7
  tags:
8
  - biology
9
  - image
10
  - animals
11
  - species
12
  - taxonomy
13
- - morphology
14
  - rare species
15
  - endangered species
16
- - data deficient species
17
- pretty_name: Rare Species Dataset
 
 
 
 
 
18
  ---
19
 
20
  # Dataset Card for Rare Species Dataset
21
 
22
  ## Dataset Description
23
 
24
- - **Homepage:**
25
- - **Repository:**
26
- - **Paper:**
27
- - **Leaderboard:**
28
- - **Point of Contact:**
29
 
30
  ### Dataset Summary
31
 
32
- This dataset uses information from [IUCN Redlist](https://www.iucnredlist.org/) to determine species that are rare (eg., endangered, critically endangered, extinct in the wild, data deficient). From this information, data (images and text) are pulled from other datasets (eg., [EOL](https://huggingface.co/datasets/imageomics/eol/)) to generate a dataset consisting of rare species for zero-shot-classification and more refined image classification tasks.
33
- There are approximately 25,000 species that fall into these categories, though the image availability for each species is yet to be determined.
34
- <!--This dataset card aims to be a base template for new datasets. It has been generated using [this raw template](https://github.com/huggingface/huggingface_hub/blob/main/src/huggingface_hub/templates/datasetcard_template.md?plain=1).-->
 
 
 
 
 
 
 
35
 
36
  ### Supported Tasks and Leaderboards
37
 
38
- [More Information Needed]
39
 
40
- ### Languages
41
 
42
- [More Information Needed]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
43
 
44
  ## Dataset Structure
45
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
46
  ### Data Instances
47
 
48
- [More Information Needed]
 
 
 
 
 
 
 
 
 
 
 
 
49
 
50
  ### Data Fields
51
 
52
- [More Information Needed]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
53
 
54
  ### Data Splits
55
 
56
- [More Information Needed]
57
 
58
  ## Dataset Creation
59
 
60
  ### Curation Rationale
61
 
62
- This dataset was generated with the purpose of providing a test set for the [Imageomics BioCLIP model](https://huggingface.co/imageomics/bioclip) to demonstrate robustness on data with minimal training samples available.
63
 
64
  ### Source Data
65
 
 
 
66
  #### Initial Data Collection and Normalization
67
 
68
- [More Information Needed]
 
 
69
 
70
- #### Who are the source language producers?
71
 
72
- [More Information Needed]
73
 
74
  ### Annotations
75
 
76
  #### Annotation process
77
 
78
- [More Information Needed]
 
 
79
 
80
  #### Who are the annotators?
81
 
82
- [More Information Needed]
83
 
84
  ### Personal and Sensitive Information
85
 
86
- [More Information Needed]
87
 
88
  ## Considerations for Using the Data
89
-
90
  ### Social Impact of Dataset
91
 
92
- [More Information Needed]
93
 
94
  ### Discussion of Biases
95
 
96
- [More Information Needed]
97
-
98
- ### Other Known Limitations
99
 
100
- [More Information Needed]
101
 
102
  ## Additional Information
103
 
104
  ### Dataset Curators
105
 
106
- [More Information Needed]
107
-
108
 
109
  ### Licensing Information
110
 
111
- The data (images and text) contain a variety of licensing restrictions mostly within the CC family.
112
- Each image and text in this dataset is provided under the least restrictive terms allowed by its licensing requirements as provided to us (i.e, we impose no additional restrictions past those specified by licenses in the license file).
 
 
 
113
 
114
  ### Citation Information
115
 
116
- [More Information Needed]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
117
 
118
- Please be sure to also cite the original data source(s) and all constituent parts as appropriate.
119
 
120
  ### Contributions
121
 
122
- The [Imageomics Institute](https://imageomics.org) is funded by the US National Science Foundation's Harnessing the Data Revolution (HDR) Institute program under [Award #2118240](https://www.nsf.gov/awardsearch/showAward?AWD_ID=2118240) (Imageomics: A New Frontier of Biological Information Powered by Knowledge-Guided Machine Learning).
 
1
  ---
2
+ License: cc0-1.0
 
 
3
  language:
4
  - en
5
+ - la
6
+ pretty_name: Rare Species Dataset
7
+ task_categories:
8
+ - image-classification
9
+ - zero-shot-classification
10
  tags:
11
  - biology
12
  - image
13
  - animals
14
  - species
15
  - taxonomy
 
16
  - rare species
17
  - endangered species
18
+ - evolutionary biology
19
+ - balanced
20
+ - CV
21
+ - multimodal
22
+ - CLIP
23
+ - knowledge-guided
24
+ size_categories: 10K<n<100K
25
  ---
26
 
27
  # Dataset Card for Rare Species Dataset
28
 
29
  ## Dataset Description
30
 
31
+ <!-- - **Homepage:** -->
32
+ - **Repository:** [Imageomics/bioclip](https://github.com/Imageomics/bioclip)
33
+ - **Paper:** BioCLIP: A Vision Foundation Model for the Tree of Life ([arXiv](https://doi.org/10.48550/arXiv.2311.18803))
34
+ <!-- - **Leaderboard:** -->
 
35
 
36
  ### Dataset Summary
37
 
38
+ This dataset was generated alongside [TreeOfLife-10M](https://huggingface.co/datasets/imageomics/TreeOfLife-10M); data (images and text) were pulled from [Encyclopedia of Life (EOL)](https://eol.org) to generate a dataset consisting of rare species for zero-shot-classification and more refined image classification tasks. Here, we use "rare species" to mean species listed on [The International Union for Conservation of Nature (IUCN) Red List](https://www.iucnredlist.org/) as Near Threatened, Vulnerable, Endangered, Critically Endangered, and Extinct in the Wild.
39
+
40
+
41
+ <!--This dataset card aims to be a base template for new datasets. It has been generated using [this raw template](https://github.com/huggingface/huggingface_hub/blob/main/src/huggingface_hub/templates/datasetcard_template.md?plain=1). And further altered to suit Imageomics Institute needs.-->
42
+
43
+
44
+ |![treemap from phyla down to family](https://huggingface.co/datasets/imageomics/rare-species/resolve/main/visuals/phyla_ToL_tree.png)|
45
+ |:--|
46
+ |**Figure 1.** Treemap from phyla down to family for Rare Species dataset. Interactive version available in [`visuals`](https://huggingface.co/imageomics/rare-species/tree/main/visuals) folder.|
47
+
48
 
49
  ### Supported Tasks and Leaderboards
50
 
51
+ Image Classification, Zero-shot and few-shot Classification.
52
 
53
+ Baseline for Random guessing is 0.3.
54
 
55
+
56
+ | Model | | Rare Species Classification Results | |
57
+ | ---- | :----: | :----: | :----: |
58
+ | | _Zero-Shot Classification_ | _One-Shot Classification_ | _Five-Shot Classification_ |
59
+ | CLIP | 31.81 | 29.26 | 47.45 |
60
+ | OpenCLIP | 29.85 | 28.52 | 46.07 |
61
+ | BioCLIP | **38.09** | **44.9** | **65.7** |
62
+ | --iNat21 Only | 21.33 | 36.94 | 55.65 |
63
+
64
+ | |
65
+ | -- |
66
+ | Zero-, one- and five-shot classification top-1 accuracy for different CLIP models. **Bold** indicates best accuracy. All models use the same architecture: ViT-B/16 vision encoders, 77-token text encoder. "iNat21 Only" follows the same procedure as BioCLIP but uses iNat21 instead of TreeOfLife-10M. CLIP and OpenCLIP are tested on common name, while BioCLIP and iNat21 Only were tested on full taxonomic name + common name. In this manner, we compare the optimal CLIP and OpenCLIP performance (both were primarily trained with common names). |
67
+
68
+ ### Languages
69
+ English, Latin
70
 
71
  ## Dataset Structure
72
 
73
+ ```
74
+ /dataset/
75
+ <kingdom-phylum-class-order-family-genus-species-1>/
76
+ <eol_content_id_1>_<eol_page_id>_eol_full-size-copy.jpg
77
+ <eol_content_id_2>_<eol_page_id>_eol_full-size-copy.jpg
78
+ ...
79
+ <eol_content_id_30>_<eol_page_id>_eol_full-size-copy.jpg
80
+ <kingdom-phylum-class-order-family-genus-species-2>/
81
+ <eol_content_id_1>_<eol_page_id>_eol_full-size-copy.jpg
82
+ <eol_content_id_2>_<eol_page_id>_eol_full-size-copy.jpg
83
+ ...
84
+ <eol_content_id_30>_<eol_page_id>_eol_full-size-copy.jpg
85
+ ...
86
+ <kingdom-phylum-class-order-family-genus-species-400>/
87
+ <eol_content_id_1>_<eol_page_id>_eol_full-size-copy.jpg
88
+ <eol_content_id_2>_<eol_page_id>_eol_full-size-copy.jpg
89
+ ...
90
+ <eol_content_id_30>_<eol_page_id>_eol_full-size-copy.jpg
91
+ metadata/
92
+ rarespecies-catalog.csv
93
+ licenses.csv
94
+ visuals/
95
+ phyla_ToL_tree.html
96
+ phyla_ToL_tree.pdf
97
+ phyla_ToL_tree.png
98
+ ```
99
+
100
  ### Data Instances
101
 
102
+ This dataset is a collection of images with associated text. The text matched to images contains both [Linnaean taxonomy](https://www.britannica.com/science/taxonomy/The-objectives-of-biological-classification) (kingdom through species) for the particular subject of the image and its scientific name (`<genus> <species>`). All images have full 7-rank taxonomy filled, and are included in the [IUCN Red List](https://www.iucnredlist.org/) categories Near Threatened, Vulnerable, Endangered, Critically Endangered, and Extinct in the Wild. There are 30 images per species for the 400 species included.*
103
+
104
+ The images in this dataset are JPGs with filenames `<eol_content_id>_<eol_page_id>_eol_full-size-copy.jpg`. See Metadata Files below for definition of the IDs.
105
+
106
+ *It was discovered after training on TreeOfLife-10M that of the 400 species held out, 5 did not actually have 30 unique images, despite each image having unique EOL content IDs and EOL full-size image URLs. These species are as follows:
107
+ | Species | Number of Unique Images |
108
+ | --- | -- |
109
+ | _Pheidole elecebra_ | 21 |
110
+ | _Calumma ambreense_ | 27 |
111
+ | _Acanthochelys macrocephala_ | 27 |
112
+ | _Haliaeetus vociferoides_ | 29 |
113
+ | _Wallago attu_ | 29 |
114
+
115
 
116
  ### Data Fields
117
 
118
+ #### Metadata Files
119
+
120
+ `rarespecies-catalog.csv`: contains the following metadata associated with each image in the dataset
121
+ - `rarespecies_id`: unique identifier for the image in the dataset.
122
+ - `eol_content_id`: unique identifier within EOL database for images sourced from [EOL](https://eol.org). Note that EOL content IDs are not stable.
123
+ - `eol_page_id`: identifier of page from which images from EOL are sourced. Note that an image's association to a particular page ID may change with updates to the EOL (or image provider's) hierarchy. However, EOL taxon page IDs are stable.
124
+ The remaining terms describe the Linnaean taxonomy of the subject of the images; application of these labels is described below in the [annotation process](#annotation-process).
125
+ - `kingdom`: kingdom to which the subject of the image belongs (all `Animalia`).
126
+ - `phylum`: phylum to which the subject of the image belongs.
127
+ - `class`: class to which the subject of the image belongs.
128
+ - `order`: order to which the subject of the image belongs.
129
+ - `family`: family to which the subject of the image belongs.
130
+ - `genus`: genus to which the subject of the image belongs.
131
+ - `species`: species to which the subject of the image belongs.
132
+ - `sciName`: scientific name associated with the subject of the image (`genus-species`).
133
+ - `common`: common name associated with the subject of the image. Note that there are only 398 unique common names; it is not uncommon for species of the same genera to share a common name. The two specific instances are _Acropora acuminata_ and _Acropora millepora_, which share the common name staghorn coral, and both _Tylototriton shanjing_ and _Tylototriton verrucosus_ have the common name Yunnan Newt.
134
+
135
+ `licenses.csv`: File with license, source, and copyright holder associated to each image listed in `rarespecies-catalog.csv`; `rarespecies_id` is the shared unique identifier to link the two files. Columns are
136
+ - `rarespecies_id`, `eol_content_id`, and `eol_page_id` are as defined above.
137
+ - `md5`: MD5 hash of the image.
138
+ - `medium_source_url`: URL pointing to source of image.
139
+ - `eol_full_size_copy_url`: URL to access the full-sized image; this is the URL from which the image was downloaded for this dataset (see [Initial Data Collection and Normalization](#initial-data-collection-and-normalization) for more information on this process).
140
+ - `license_name`: name of license attached to the image (eg., `cc-by`).
141
+ - `copyright_owner`: copyright holder for the image, filled with `not provided` if no copyright owner was provided.
142
+ - `license_link`: URL to the listed license, left null in the case that `License Name` is `No known copyright restrictions`.
143
+ - `title`: title provided for the image, filled with `not provided` if no title was provided.
144
+
145
+
146
+
147
+ The visuals folder has treemaps that were generated by feeding `rarespecies-catalog.csv` to the `taxa_viz` script in the [BioCLIP GitHub repository](https://github.com/Imageomics/bioclip).
148
+
149
 
150
  ### Data Splits
151
 
152
+ This entire dataset was used for testing the [BioCLIP model](https://huggingface.co/imageomics/bioclip), which was trained on [TreeOfLife-10M](https://huggingface.co/datasets/imageomics/TreeOfLife-10M).
153
 
154
  ## Dataset Creation
155
 
156
  ### Curation Rationale
157
 
158
+ This dataset was generated with the purpose of providing a biologically meaningful test set for the [Imageomics BioCLIP model](https://huggingface.co/imageomics/bioclip) to demonstrate robustness on data with minimal training samples available and biologically meaningful potential applications.
159
 
160
  ### Source Data
161
 
162
+ [EOL](https://eol.org) and [IUCN Red List](https://www.iucnredlist.org/)
163
+
164
  #### Initial Data Collection and Normalization
165
 
166
+ The IUCN Red List of Threatened Species categorization of animals was pulled from the [IUCN website](https://www.iucnredlist.org/). There are approximately 25,000 species that fall into the categories Near Threatened, Vulnerable, Endangered, Critically Endangered, and Extinct in the Wild (as of July 13, 2023), though image availability on EOL is not consistent across species. We select 400 species from the list under the condition there are at least 30 images per species available and they are not species in [iNat21](https://kaggle.com/competitions/inaturalist-2021) or [BIOSCAN-1M](https://zenodo.org/doi/10.5281/zenodo.8030064) datasets which were also used to generate [TreeOfLife-10M](https://huggingface.co/datasets/imageomics/TreeOfLife-10M). A random subset of 30 images is then selected for each species in this collection.
167
+
168
+ This dataset was generated concurrently with [TreeOfLife-10M](https://huggingface.co/datasets/imageomics/TreeOfLife-10M), so the process is as described [there](https://huggingface.co/datasets/imageomics/TreeOfLife-10M#initial-data-collection-and-normalization), with the exception that these images were entirely sourced from EOL, and the species represented were excluded from the TreeOfLife-10M dataset.
169
 
170
+ The IUCN data was used for selection of the included species, and is not reproduced here. [This link](https://www.iucnredlist.org/search?permalink=ab8daad6-d564-4370-b8e6-9c5ac9f8336f) provides the search used to gather the list of species classified as Near Threatened to Extinct in the Wild. The results were downloaded on July 13, 2023, but note the results are subject to change with IUCN Red List Updates ([IUCN Update Schedule](https://www.iucnredlist.org/assessment/updates)).
171
 
 
172
 
173
  ### Annotations
174
 
175
  #### Annotation process
176
 
177
+ Annotations were primarily sourced from EOL (image source provider) following the procedure described in the [TreeOfLife-10M annotation process](https://huggingface.co/datasets/imageomics/TreeOfLife-10M#annotation-process). [IUCN Red List](https://www.iucnredlist.org/) was then used for filtering these taxa out of [TreeOfLife-10M](https://huggingface.co/datasets/imageomics/TreeOfLife-10M) to create this Rare Species dataset.
178
+
179
+ The scientific name (`genus-species`, as labeled by EOL) was used to look up the higher-order taxa from EOL aggregate datasets (described below), then matched against the ITIS hierarchy for the higher-order taxa standardization. A small number of these are [homonyms](https://en.wikipedia.org/wiki/Homonym_(biology)), for which a list was generated to ensure proper matching of higher-order taxa. After these resources were exhausted, any remaining unresolved taxa were fed through the [Global Names Resolver (GNR) API](https://resolver.globalnames.org/api).
180
 
181
  #### Who are the annotators?
182
 
183
+ Samuel Stevens, Jiaman Wu, Matthew J. Thompson, and Elizabeth G. Campolongo
184
 
185
  ### Personal and Sensitive Information
186
 
187
+ All animals included in this dataset are listed as Near Threatened, Vulnerable, Endangered, Critically Endangered, or Extinct in the Wild by the [IUCN Red List](https://www.iucnredlist.org/) as of July 13, 2023 (IUCN generally updates classifications twice each year (see the [IUCN Update Schedule](https://www.iucnredlist.org/assessment/updates) for more information)). However, the specific ranking is not tied to any individual, and there is no geographical information included.
188
 
189
  ## Considerations for Using the Data
 
190
  ### Social Impact of Dataset
191
 
192
+ The hope is that this dataset could be helpful in conservation efforts or biodiversity research.
193
 
194
  ### Discussion of Biases
195
 
196
+ Inclusion of a species in this dataset required that EOL provided at least 30 images of it, so there are only 400 of the 25,000 species in these categories included, and only 30 images per species. Additionally, all included species are in the kingdom, _Animalia_, and within 5 phyla.
 
 
197
 
 
198
 
199
  ## Additional Information
200
 
201
  ### Dataset Curators
202
 
203
+ Samuel Stevens, Jiaman Wu, Matthew J. Thompson, and Elizabeth G. Campolongo
 
204
 
205
  ### Licensing Information
206
 
207
+ The data (images and text) contain a variety of licensing restrictions ranging from [CC0](https://creativecommons.org/publicdomain/zero/1.0/) to [CC BY-NC-SA](https://creativecommons.org/licenses/by-nc-sa/4.0/). Each image and text in this dataset is provided under the least restrictive terms allowed by its licensing requirements as provided to us (i.e, we impose no additional restrictions past those specified by licenses in the license file).
208
+
209
+ This dataset (the compilation) has been marked as dedicated to the public domain by applying the [CC0 Public Domain Waiver](https://creativecommons.org/publicdomain/zero/1.0/). However, images may be licensed under different terms (as noted above).
210
+ For license and citation information by image, see our [license file](https://huggingface.co/datasets/imageomics/rare-species/blob/main/metadata/licenses.csv).
211
+
212
 
213
  ### Citation Information
214
 
215
+ ```
216
+ @dataset{rare_species_2023,
217
+ author = {Samuel Stevens and Jiaman Wu and Matthew J Thompson and Elizabeth G Campolongo and Chan Hee Song and David Edward Carlyn and Li Dong and Wasila M Dahdul and Charles Stewart and Tanya Berger-Wolf and Wei-Lun Chao and Yu Su},
218
+ title = {Rare Species},
219
+ year = {2023},
220
+ url = {https://huggingface.co/datasets/imageomics/rare-species},
221
+ doi = {<doi once generated>},
222
+ publisher = {Hugging Face}
223
+ }
224
+ ```
225
+ Please also cite our paper:
226
+ ```
227
+ @article{stevens2023bioclip,
228
+ title = {BIOCLIP: A Vision Foundation Model for the Tree of Life},
229
+ author = {Samuel Stevens and Jiaman Wu and Matthew J Thompson and Elizabeth G Campolongo and Chan Hee Song and David Edward Carlyn and Li Dong and Wasila M Dahdul and Charles Stewart and Tanya Berger-Wolf and Wei-Lun Chao and Yu Su},
230
+ year = {2023},
231
+ eprint = {2311.18803},
232
+ archivePrefix = {arXiv},
233
+ primaryClass = {cs.CV}}
234
+ ```
235
+
236
+
237
+ Please be sure to also cite the original data sources and all constituent parts as appropriate.
238
+
239
+
240
+ **EOL and IUCN classification data:**
241
+
242
+ IUCN. 2022. The IUCN Red List of Threatened Species. Version 2022-2. https://www.iucnredlist.org. Accessed on 5 July 2023. https://www.iucnredlist.org/search?permalink=ab8daad6-d564-4370-b8e6-9c5ac9f8336f.
243
+
244
+ Encyclopedia of Life. Available from http://eol.org. Accessed 29 July 2023.
245
+
246
+
247
+ For license and citation information by image, see our [license file](https://huggingface.co/datasets/imageomics/rare-species/blob/main/metadata/licenses.csv).
248
 
 
249
 
250
  ### Contributions
251
 
252
+ The [Imageomics Institute](https://imageomics.org) is funded by the US National Science Foundation's Harnessing the Data Revolution (HDR) program under [Award #2118240](https://www.nsf.gov/awardsearch/showAward?AWD_ID=2118240) (Imageomics: A New Frontier of Biological Information Powered by Knowledge-Guided Machine Learning). Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.