gremlin97 commited on
Commit
16bbebd
·
verified ·
1 Parent(s): 6cda309

Add dataset README with metadata

Browse files
Files changed (1) hide show
  1. README.md +127 -59
README.md CHANGED
@@ -1,61 +1,129 @@
1
  ---
2
- dataset_info:
3
- features:
4
- - name: image
5
- dtype: string
6
- - name: labels
7
- sequence: int64
8
- - name: feature_names
9
- sequence: string
10
- splits:
11
- - name: train
12
- num_bytes: 677849
13
- num_examples: 1762
14
- - name: val
15
- num_bytes: 169357
16
- num_examples: 443
17
- - name: test
18
- num_bytes: 283502
19
- num_examples: 739
20
- - name: few_shot_train_10_shot
21
- num_bytes: 504166
22
- num_examples: 1310
23
- - name: few_shot_train_15_shot
24
- num_bytes: 522251
25
- num_examples: 1357
26
- - name: few_shot_train_1_shot
27
- num_bytes: 460626
28
- num_examples: 1198
29
- - name: few_shot_train_20_shot
30
- num_bytes: 539801
31
- num_examples: 1402
32
- - name: few_shot_train_2_shot
33
- num_bytes: 466086
34
- num_examples: 1212
35
- - name: few_shot_train_5_shot
36
- num_bytes: 480488
37
- num_examples: 1249
38
- download_size: 330295
39
- dataset_size: 4104126
40
- configs:
41
- - config_name: default
42
- data_files:
43
- - split: train
44
- path: data/train-*
45
- - split: val
46
- path: data/val-*
47
- - split: test
48
- path: data/test-*
49
- - split: few_shot_train_10_shot
50
- path: data/few_shot_train_10_shot-*
51
- - split: few_shot_train_15_shot
52
- path: data/few_shot_train_15_shot-*
53
- - split: few_shot_train_1_shot
54
- path: data/few_shot_train_1_shot-*
55
- - split: few_shot_train_20_shot
56
- path: data/few_shot_train_20_shot-*
57
- - split: few_shot_train_2_shot
58
- path: data/few_shot_train_2_shot-*
59
- - split: few_shot_train_5_shot
60
- path: data/few_shot_train_5_shot-*
61
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ annotations_creators:
3
+ - expert-generated
4
+ language_creators:
5
+ - found
6
+ language:
7
+ - en
8
+ license:
9
+ - cc-by-4.0
10
+ multilinguality:
11
+ - monolingual
12
+ size_categories:
13
+ - 1K<n<10K
14
+ source_datasets:
15
+ - original
16
+ task_categories:
17
+ - image-classification
18
+ task_ids:
19
+ - multi-label-image-classification
20
+ pretty_name: mars-multi-label-classification
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
21
  ---
22
+
23
+ # mars-multi-label-classification
24
+
25
+ A Mars image multi-label classification dataset for planetary science research.
26
+
27
+ ## Dataset Metadata
28
+
29
+ * **License:** CC-BY-4.0 (Creative Commons Attribution 4.0 International)
30
+ * **Version:** 1.0
31
+ * **Date Published:** 2025-05-10
32
+ * **Cite As:** TBD
33
+
34
+ ## Classes
35
+
36
+ This dataset uses multi-label classification, meaning each image can have multiple class labels.
37
+
38
+ The dataset contains the following classes:
39
+
40
+ - **rah** (0): Rock Abrasion Tool (RAT) Hole
41
+ - **cla** (1): Clasts
42
+ - **dur** (2): Dunes/Ripples
43
+ - **soi** (3): Soil
44
+ - **roc** (4): Rock Outcrops
45
+ - **clr** (5): Close-up Rock
46
+ - **rab** (6): Rock Abrasion Tool (RAT) Brushed Target
47
+ - **div** (7): Distant Vista
48
+ - **rod** (8): Rover Deck
49
+ - **bso** (9): Bright Soil
50
+ - **flr** (10): Float Rocks
51
+ - **art** (11): Artifacts
52
+ - **pct** (12): Pancam Calibration Target
53
+ - **arh** (13): Arm Hardware
54
+ - **rrf** (14): Rock (Round Features)
55
+ - **sph** (15): Spherules
56
+ - **ohw** (16): Other Hardware
57
+ - **ast** (17): Astronomy
58
+ - **nbs** (18): Nearby Surface
59
+ - **rmi** (19): Rocks (Misc)
60
+ - **rtr** (20): Rover Tracks
61
+ - **sky** (21): Sky
62
+ - **rpa** (22): Rover Parts
63
+ - **rlf** (23): Rock (Linear Features)
64
+ - **sot** (24): Soil Trench
65
+ ## Statistics
66
+
67
+ - **train**: 1762 images
68
+ - **val**: 443 images
69
+ - **test**: 739 images
70
+ - **few_shot_train_10_shot**: 1310 images
71
+ - **few_shot_train_15_shot**: 1357 images
72
+ - **few_shot_train_1_shot**: 1198 images
73
+ - **few_shot_train_20_shot**: 1402 images
74
+ - **few_shot_train_2_shot**: 1212 images
75
+ - **few_shot_train_5_shot**: 1249 images
76
+
77
+ ## Few-shot Splits
78
+
79
+ This dataset includes the following few-shot training splits:
80
+
81
+ - **few_shot_train_10_shot**: 1310 images
82
+ - **few_shot_train_15_shot**: 1357 images
83
+ - **few_shot_train_1_shot**: 1198 images
84
+ - **few_shot_train_20_shot**: 1402 images
85
+ - **few_shot_train_2_shot**: 1212 images
86
+ - **few_shot_train_5_shot**: 1249 images
87
+
88
+ Few-shot configurations:
89
+
90
+ - **10_shot.csv**
91
+ - **15_shot.csv**
92
+ - **1_shot.csv**
93
+ - **20_shot.csv**
94
+ - **2_shot.csv**
95
+ - **5_shot.csv**
96
+ ## Format
97
+
98
+ Each example in the dataset has the following format:
99
+
100
+ ```
101
+ {
102
+ 'image': Image(...), # PIL image
103
+ 'labels': List[int], # Multi-hot encoded binary vector (1 if class is present, 0 otherwise)
104
+ 'feature_names': List[str], # List of feature names (class short codes)
105
+ }
106
+ ```
107
+
108
+ ## Usage
109
+
110
+ ```python
111
+ from datasets import load_dataset
112
+
113
+ dataset = load_dataset("gremlin97/mars-multi-label-classification")
114
+
115
+ # Access an example
116
+ example = dataset['train'][0]
117
+ image = example['image'] # PIL image
118
+ labels = example['labels'] # Multi-hot encoded binary vector
119
+
120
+ # Example of how to find which classes are present in an image
121
+ present_classes = [i for i, is_present in enumerate(labels) if is_present == 1]
122
+ print(f"Classes present in this image: {present_classes}")
123
+ ```
124
+
125
+ ## Multi-label Classification
126
+
127
+ In multi-label classification, each image can belong to multiple classes simultaneously. The labels are represented as a binary vector where a 1 indicates the presence of a class and a 0 indicates its absence.
128
+
129
+ Unlike single-label classification where each image has exactly one class, multi-label classification allows modeling scenarios where multiple features can be present in the same image, which is often the case with Mars imagery.