Update README.md
Browse files
README.md
CHANGED
@@ -48,32 +48,6 @@ dataset_split_names:
|
|
48 |
As part of an ongoing worldwide effort to comprehend and monitor insect biodiversity, we present the BIOSCAN-5M Insect dataset to the machine learning community. BIOSCAN-5M is a comprehensive dataset containing multi-modal information for over 5 million insect specimens,
|
49 |
and it significantly expands existing image-based biological datasets by including taxonomic labels, raw nucleotide barcode sequences, assigned barcode index numbers, geographical information, and specimen size.
|
50 |
|
51 |
-
### Large-Scale Foundation Model Training
|
52 |
-
|
53 |
-
#### BIOSCAN-5M is partitioned to support both closed-world and open-world biodiversity learning:
|
54 |
-
|
55 |
-
- **Closed-world (train, val, test):**
|
56 |
-
Samples with known species names for supervised classification.
|
57 |
-
|
58 |
-
- **Open-world (key_unseen, val_unseen, test_unseen):**
|
59 |
-
Placeholder species but known genera, enabling generalization to unseen species.
|
60 |
-
|
61 |
-
- **Novelty detection (other_heldout):**
|
62 |
-
Unknown species and genus, suitable for open-set detection.
|
63 |
-
|
64 |
-
- **Pretraining (pretrain):**
|
65 |
-
Unlabeled data for self- or semi-supervised learning at scale.
|
66 |
-
|
67 |
-
#### Supported Tasks
|
68 |
-
- **Task I: DNA-based Taxonomic Classification**
|
69 |
-
Predict taxonomic labels from raw DNA barcode sequences.
|
70 |
-
|
71 |
-
- **Task II: Zero-Shot Transfer Learning**
|
72 |
-
Evaluate whether unlabeled models can organize data into semantically meaningful clusters—across modalities like image and DNA—using learned representations.
|
73 |
-
|
74 |
-
- **Task III: Multimodal Retrieval Learning**
|
75 |
-
Retrieve matching specimens across modalities (e.g., image ↔ DNA ↔ Text) using shared embeddings.
|
76 |
-
|
77 |
### Citation
|
78 |
|
79 |
If you make use of the BIOSCAN-5M dataset and/or its code repository, please cite the following paper:
|
@@ -98,6 +72,26 @@ cite as:
|
|
98 |
}
|
99 |
```
|
100 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
101 |
|
102 |
## Dataset features via metadata fields
|
103 |
|
|
|
48 |
As part of an ongoing worldwide effort to comprehend and monitor insect biodiversity, we present the BIOSCAN-5M Insect dataset to the machine learning community. BIOSCAN-5M is a comprehensive dataset containing multi-modal information for over 5 million insect specimens,
|
49 |
and it significantly expands existing image-based biological datasets by including taxonomic labels, raw nucleotide barcode sequences, assigned barcode index numbers, geographical information, and specimen size.
|
50 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
51 |
### Citation
|
52 |
|
53 |
If you make use of the BIOSCAN-5M dataset and/or its code repository, please cite the following paper:
|
|
|
72 |
}
|
73 |
```
|
74 |
|
75 |
+
## Large-Scale Foundation Model Training
|
76 |
+
|
77 |
+
#### Dataset Partitions
|
78 |
+
|
79 |
+
| Partition | Description |
|
80 |
+
|------------------|-----------------------------------------------------------------------------|
|
81 |
+
| **Closed-world**<br>(train, val, test) | Samples with known species names for supervised classification. |
|
82 |
+
| **Open-world**<br>(key_unseen, val_unseen, test_unseen) | Placeholder species names but known genera, enabling generalization to unseen species. |
|
83 |
+
| **Novelty Detection**<br>(other_heldout) | Unknown species and genus, suitable for open-set detection. |
|
84 |
+
| **Pretraining**<br>(pretrain) | Unlabeled data for self-/semi-supervised learning at scale. |
|
85 |
+
|
86 |
+
|
87 |
+
#### Supported Tasks
|
88 |
+
|
89 |
+
| Task | Description |
|
90 |
+
|-------------------------------|-----------------------------------------------------------------------------|
|
91 |
+
| **Task I**<br>DNA-based Taxonomic Classification | Predict taxonomic labels from raw DNA barcode sequences. |
|
92 |
+
| **Task II**<br>Zero-Shot Transfer Learning | Evaluate whether unlabeled models can semantically cluster data—across modalities like image and DNA—using learned representations. |
|
93 |
+
| **Task III**<br>Multimodal Retrieval Learning | Retrieve matching specimens across modalities (e.g., image ↔ DNA ↔ text) via shared embeddings. |
|
94 |
+
|
95 |
|
96 |
## Dataset features via metadata fields
|
97 |
|