astronolan commited on
Commit
3bd6c3d
·
verified ·
1 Parent(s): 2adeabe

Upload Galaxy10 AION-1 benchmark dataset (train/test split)

Browse files
README.md ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ pretty_name: "Galaxy10 AION-1 Benchmark"
3
+ license: mit
4
+
5
+ tags:
6
+ - astronomy
7
+ - galaxy-morphology
8
+ - aion
9
+ - galaxy-zoo
10
+
11
+ size_categories:
12
+ - 1K<n<10K
13
+
14
+ configs:
15
+ - config_name: default
16
+ data_files:
17
+ - split: train
18
+ path: data/train-*.parquet
19
+ - split: test
20
+ path: data/test-*.parquet
21
+
22
+ dataset_info:
23
+ features:
24
+ - name: image_rgb
25
+ dtype: image
26
+ - name: ra
27
+ dtype: float64
28
+ - name: dec
29
+ dtype: float64
30
+ - name: Galaxy10_DECals_index
31
+ dtype: int64
32
+ - name: label
33
+ dtype: int64
34
+ - name: label_name
35
+ dtype: string
36
+ - name: image_bands
37
+ dtype: binary
38
+ ---
39
+
40
+ # Galaxy10 AION-1 Benchmark
41
+
42
+ This dataset provides the exact train/test split used to produce the Galaxy Morphology Classification results in the [AION-1 paper](https://arxiv.org/abs/2510.17960) (Table 2, Section 7.2.2) and used in the [AION-Search paper](https://arxiv.org/abs/2512.11982).
43
+
44
+ ## Task
45
+
46
+ Classify galaxy images into 10 morphology classes from [Galaxy Zoo DECaLS](https://ui.adsabs.harvard.edu/link_gateway/2008MNRAS.389.1179L/PUB_HTML):
47
+
48
+ | Label | Class Name |
49
+ |-------|-----------|
50
+ | 0 | Disturbed Galaxies |
51
+ | 1 | Merging Galaxies |
52
+ | 2 | Round Smooth Galaxies |
53
+ | 3 | In-between Round Smooth Galaxies |
54
+ | 4 | Cigar Shaped Smooth Galaxies |
55
+ | 5 | Barred Spiral Galaxies |
56
+ | 6 | Unbarred Tight Spiral Galaxies |
57
+ | 7 | Unbarred Loose Spiral Galaxies |
58
+ | 8 | Edge-on Galaxies without Bulge |
59
+ | 9 | Edge-on Galaxies with Bulge |
60
+
61
+ ## Splits
62
+
63
+ | Split | Count | Description |
64
+ |-------|-------|-------------|
65
+ | train | 7,120 | 90% class-stratified split |
66
+ | test | 796 | 10% held-out evaluation set |
67
+
68
+ These are the exact indices used in the AION-1 paper. The test set produces the accuracy numbers in Table 2.
69
+
70
+ ## Columns
71
+
72
+ | Column | Type | Description |
73
+ |--------|------|-------------|
74
+ | `image_rgb` | image | 256x256x3 uint8 RGB from Galaxy10_DECals.h5 (PNG) |
75
+ | `ra` | float64 | Right ascension (degrees) |
76
+ | `dec` | float64 | Declination (degrees) |
77
+ | `Galaxy10_DECals_index` | int64 | Row index into Galaxy10_DECals.h5 |
78
+ | `label` | int64 | Morphology class (0-9) |
79
+ | `label_name` | string | Human-readable class name |
80
+ | `image_bands` | binary | 96x96 4-band (g,r,i,z) float32 cutout from Legacy Survey DR10 |
81
+
82
+ ### Reconstructing `image_bands`
83
+
84
+ The `image_bands` column stores raw float32 flux values as bytes. To reconstruct:
85
+
86
+ ```python
87
+ import numpy as np
88
+ cutout = np.frombuffer(row["image_bands"], dtype=np.float32).reshape(4, 96, 96)
89
+ # cutout[0] = g-band, cutout[1] = r-band, cutout[2] = i-band, cutout[3] = z-band
90
+ ```
91
+
92
+ ### Using with AION-1
93
+
94
+ To tokenize with the AION codec and compute embeddings:
95
+
96
+ ```python
97
+ import torch
98
+ from aion.codecs import CodecManager
99
+ from aion.modalities import LegacySurveyImage
100
+ from aion.model import AION
101
+
102
+ codec_manager = CodecManager(device="cuda")
103
+ model = AION.from_pretrained("polymathic-ai/aion-base").to("cuda").eval()
104
+
105
+ cutout = np.frombuffer(row["image_bands"], dtype=np.float32).reshape(4, 96, 96)
106
+ image_flux = torch.tensor(cutout).unsqueeze(0).to("cuda")
107
+ image = LegacySurveyImage(flux=image_flux, bands=["DES-G", "DES-R", "DES-I", "DES-Z"])
108
+ tokens = codec_manager.encode(image)
109
+ embeddings = model.encode({"tok_image": tokens["tok_image"]}, num_encoder_tokens=600)
110
+ mean_embedding = embeddings.mean(dim=1) # (1, 768)
111
+ ```
112
+
113
+ ## AION-1 Paper Results (Table 2)
114
+
115
+ | Model | Accuracy (%) |
116
+ |-------|-------------|
117
+ | AION-B | 84.0 |
118
+ | AION-L | 87.2 |
119
+ | AION-XL | 86.5 |
120
+ | Oquab et al. (2023) | 71.4 |
121
+ | EfficientNet | 80.0 |
122
+ | Walmsley et al. (2022) | 89.6 |
123
+ | AION-Search | X |
124
+
125
+ ## Data Sources
126
+
127
+ - **Galaxy10_DECals.h5**: Galaxy Zoo 10 catalog ([Leung and Bovy, 2018](https://astronn.readthedocs.io/en/latest/galaxy10.html); [Walmsley et al., 2022](https://academic.oup.com/mnras/article/509/3/3966/6378289))
128
+ - **4-band cutouts**: Downloaded from [ALASKY](https://aladin.cds.unistra.fr/) via CDS HiPS2FITS at 0.262 arcsec/pixel
129
+
130
+ The underlying image data is from the DESI Legacy Imaging Surveys and Galaxy Zoo, subject to their respective data-use policies.
data/test-00000-of-00002.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0175190b3b2b073ce26532e975a3ddb777a8efe09db2777fb407016efff21993
3
+ size 117153124
data/test-00001-of-00002.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f68559ba8cfb2f9693717f3008b5e9419265ae78e60e7599e906881560a349e3
3
+ size 117425679
data/train-00000-of-00010.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fc97c4ef12b14c68cab851ec75f5ae1590e1eaa8373f660129a2e328c0666efc
3
+ size 209936188
data/train-00001-of-00010.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d191a0bebea725f81eacd5e8e711cceb9b9a27aa484c37c68196f13d86a4cc00
3
+ size 208067837
data/train-00002-of-00010.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d173ecbc0cabbedffc9104fbddfb4daba9dc8175a3b06c2a793885a2a2821df1
3
+ size 210418022
data/train-00003-of-00010.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f4b58b1445a65c827e5c533b5a932446e73694231c7be32754245f6b961fcc76
3
+ size 210392978
data/train-00004-of-00010.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:124581ee8a30fbf079ecb0fba89140ffb4ced2a9c01cb01a8be065c3fd80b308
3
+ size 211009260
data/train-00005-of-00010.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:62e552dbd3662eb2f41461a82c5913c6374a69cf893eb4b8ab382eb30e7a60a4
3
+ size 208465491
data/train-00006-of-00010.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c48db1f0e6586b8ba94125af9350ea0d2d18206f0c04d571eebac8a4b37aa50f
3
+ size 209045831
data/train-00007-of-00010.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:906ea9743606b59870cf9b7d511ad7810439831d491dbd8329af14d43ffa8174
3
+ size 210667625
data/train-00008-of-00010.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7f1f1201ec8a6c20530ab4811727c3a450b2e7c47e8a0f95003bcdfecbd5152b
3
+ size 209420463
data/train-00009-of-00010.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c014b44d1135bc6776b844ef1c0c3cd04ad9f1e4326df2142f1a7bd94cd2082b
3
+ size 210384371