jaychempan commited on
Commit
83bf5ed
·
verified ·
1 Parent(s): c68a26d

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +92 -3
README.md CHANGED
@@ -1,3 +1,92 @@
1
- ---
2
- license: mit
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ task_categories:
4
+ - text-to-image
5
+ - image-to-image
6
+ - mask-generation
7
+ - image-segmentation
8
+ language:
9
+ - en
10
+ size_categories:
11
+ - 100K<n<1M
12
+ source_datasets:
13
+ - OpenEarthMap
14
+ - LoveDA
15
+ - DeepGlobe
16
+ - SAMRS
17
+ - LAE-1M
18
+ tags:
19
+ - diffusion-models
20
+ - remote-sensing
21
+ - image-synthesis
22
+ - controlnet
23
+ - earth-observation
24
+ - generative-models
25
+ pretty_name: EarthSynth-180K
26
+ ---
27
+
28
+
29
+ # EarthSynth-180K Dataset
30
+
31
+ <p align="center">
32
+ <img src="https://jianchengpan.space/EarthSynth-website/assets/EarthSynth-180K.png" alt="EarthSynth-180K" width="600"/>
33
+ </p>
34
+
35
+ **EarthSynth-180K** is a **multi-task, conditional, diffusion-based generative dataset** designed for remote sensing image synthesis and understanding.
36
+ It was introduced in the paper *"EarthSynth: Generating Informative Earth Observation with Diffusion Models"* (arXiv 2025).
37
+
38
+ This dataset supports **text-to-image generation**, **mask-conditioned synthesis**, and **multi-category augmentation** for Earth observation research.
39
+
40
+ ---
41
+
42
+ ## Dataset Details
43
+
44
+ ### Dataset Description
45
+
46
+ - **Curated by:** Jiancheng Pan, Shiye Lei, Yuqian Fu, Jiahao Li, Yanxing Liu, Yuze Sun, Xiao He, Long Peng, Xiaomeng Huang, Bo Zhao
47
+ - **Funded by:** [Not specified]
48
+ - **Shared by:** EarthSynth Team
49
+ - **Language(s):** English (for prompts)
50
+ - **License:** MIT License
51
+
52
+ ### Dataset Sources
53
+
54
+ - **Repository:** [GitHub - EarthSynth](https://github.com/jaychempan/EarthSynth)
55
+ - **Paper:** [ArXiv 2505.12108](https://arxiv.org/abs/2505.12108)
56
+ - **Project Page:** [EarthSynth Website](https://jianchengpan.space/EarthSynth-website/index.html)
57
+ - **Dataset Download:** [HuggingFace](https://huggingface.co/datasets/jaychempan/EarthSynth-180K)
58
+
59
+ ---
60
+
61
+ ## Dataset Structure
62
+
63
+ | Subset | # Images | Annotations | Format | Condition Types |
64
+ |-------------|-----------|--------------------|------------------|---------------------------|
65
+ | Train | 180,000 | Masks, Prompts | PNG + JSONL | Mask + Text |
66
+ | Validation | 10,000 | Masks, Prompts | PNG + JSONL | Mask + Text |
67
+ | Augmented | 180,000 | Single-Category | PNG + JSONL | Category + Mask + Text |
68
+
69
+ - **Masks:** Binary/instance masks for each object category.
70
+ - **Prompts:** Text prompts for conditional generation.
71
+ - **Augmentation:** Single-category augmentation for CF-Comp training strategy.
72
+
73
+ ---
74
+
75
+ ## Quick Start
76
+
77
+ ```python
78
+ from datasets import load_dataset
79
+
80
+ # Load dataset
81
+ dataset = load_dataset("jaychempan/EarthSynth-180K", split="train")
82
+
83
+ # Access one example
84
+ example = dataset[0]
85
+ print(example.keys()) # ['image', 'mask', 'prompt']
86
+
87
+ # Display image
88
+ from PIL import Image
89
+ import io
90
+
91
+ img = Image.open(io.BytesIO(example["image"]))
92
+ img.show()