Update README.md
Browse files
README.md
CHANGED
|
@@ -62,8 +62,65 @@ If you make use of the BIOSCAN-5M dataset and/or its code repository, please cit
|
|
| 62 |
}
|
| 63 |
```
|
| 64 |
|
| 65 |
-
## Dataset
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 66 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 67 |
To clone this dataset repository, use the following command:
|
| 68 |
|
| 69 |
```bash
|
|
|
|
| 62 |
}
|
| 63 |
```
|
| 64 |
|
| 65 |
+
## Dataset Structure
|
| 66 |
+
|
| 67 |
+
The dataset provides the following configurations:
|
| 68 |
+
|
| 69 |
+
- `original_256`
|
| 70 |
+
- `original_256_pretrain`
|
| 71 |
+
- `original_256_train`
|
| 72 |
+
- `original_256_eval`
|
| 73 |
+
- `cropped_256`
|
| 74 |
+
- `cropped_256_pretrain`
|
| 75 |
+
- `cropped_256_train`
|
| 76 |
+
- `cropped_256_eval`
|
| 77 |
+
|
| 78 |
+
## Dataset Features
|
| 79 |
+
|
| 80 |
+
Each configuration loads specimen images along with associated metadata fields:
|
| 81 |
+
|
| 82 |
+
- **`image`**
|
| 83 |
+
RGB JPEG images of individual insect specimens.
|
| 84 |
+
|
| 85 |
+
- **Metadata fields**
|
| 86 |
+
- **Indexing fields**
|
| 87 |
+
- `processid`: A unique number assigned by BOLD (International Barcode of Life Consortium).
|
| 88 |
+
- `sampleid`: A unique identifier given by the collector.
|
| 89 |
+
- **Taxonomic labels**
|
| 90 |
+
- `phylum`, `class`, `order`, `family`, `subfamily`, `genus`, `species`: Hierarchical taxonomic classification
|
| 91 |
+
- **Genetic information**
|
| 92 |
+
- `dna_bin`: Barcode Index Number
|
| 93 |
+
- `dna_barcode`: DNA barcode sequence
|
| 94 |
+
- **Geographic data**
|
| 95 |
+
- `country`, `province_state`: Collection location
|
| 96 |
+
- `coord-lat`, `coord-lon`: Latitude and longitude coordinates
|
| 97 |
+
- **Specimen size data**
|
| 98 |
+
- `image_measurement_value`: Measured image property (e.g., pixel area)
|
| 99 |
+
- `area_fraction`: Fraction of the image occupied by the specimen
|
| 100 |
+
- `scale_factor`: Relative size normalization factor
|
| 101 |
+
- **Split and localization**
|
| 102 |
+
- `split`: Data partition label (e.g., train, test, val)
|
| 103 |
+
- `chunk`: Subdirectory label for scalable storage
|
| 104 |
+
|
| 105 |
+
## Usage
|
| 106 |
+
|
| 107 |
+
To load the BIOSCAN-5M dataset locally from bioscan5m.py
|
| 108 |
+
|
| 109 |
+
```python
|
| 110 |
+
from datasets import load_dataset
|
| 111 |
+
|
| 112 |
+
ds = load_dataset("bioscan5m.py", name="cropped_256_eval", split="validation", trust_remote_code=True)
|
| 113 |
+
```
|
| 114 |
+
|
| 115 |
+
To load the dataset from the Hugging Face Hub:
|
| 116 |
+
|
| 117 |
+
```python
|
| 118 |
+
from datasets import load_dataset
|
| 119 |
|
| 120 |
+
dataset = load_dataset("bioscan-ml/BIOSCAN-5M", name="cropped_256_eval", split="validation")
|
| 121 |
+
```
|
| 122 |
+
|
| 123 |
+
## Dataset Access
|
| 124 |
To clone this dataset repository, use the following command:
|
| 125 |
|
| 126 |
```bash
|