edns 1.0.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- edns-1.0.0/LICENSE +21 -0
- edns-1.0.0/MANIFEST.in +8 -0
- edns-1.0.0/PKG-INFO +172 -0
- edns-1.0.0/README.md +117 -0
- edns-1.0.0/configs/data/probav.yaml +179 -0
- edns-1.0.0/configs/data/sentinel2.yaml +220 -0
- edns-1.0.0/configs/degradation/alsat2b.yaml +137 -0
- edns-1.0.0/configs/models/edsr_small.yaml +30 -0
- edns-1.0.0/configs/models/edsr_tiny.yaml +28 -0
- edns-1.0.0/configs/models/realesrgan.yaml +50 -0
- edns-1.0.0/configs/models/swinir_base.yaml +39 -0
- edns-1.0.0/pyproject.toml +110 -0
- edns-1.0.0/setup.cfg +4 -0
- edns-1.0.0/setup.py +62 -0
- edns-1.0.0/src/edns/__init__.py +60 -0
- edns-1.0.0/src/edns/configs/models/edsr_small.yaml +30 -0
- edns-1.0.0/src/edns/data/__init__.py +33 -0
- edns-1.0.0/src/edns/data/io.py +124 -0
- edns-1.0.0/src/edns/data/normalization.py +190 -0
- edns-1.0.0/src/edns/evaluation/__init__.py +22 -0
- edns-1.0.0/src/edns/evaluation/metrics.py +432 -0
- edns-1.0.0/src/edns/hallucination/__init__.py +31 -0
- edns-1.0.0/src/edns/hallucination/detection.py +732 -0
- edns-1.0.0/src/edns/inference/__init__.py +14 -0
- edns-1.0.0/src/edns/inference/app.py +335 -0
- edns-1.0.0/src/edns/models/__init__.py +13 -0
- edns-1.0.0/src/edns/models/edsr.py +218 -0
- edns-1.0.0/src/edns/utils/__init__.py +15 -0
- edns-1.0.0/src/edns/utils/helpers.py +90 -0
- edns-1.0.0/src/edns/utils/logging.py +53 -0
- edns-1.0.0/src/edns.egg-info/PKG-INFO +172 -0
- edns-1.0.0/src/edns.egg-info/SOURCES.txt +34 -0
- edns-1.0.0/src/edns.egg-info/dependency_links.txt +1 -0
- edns-1.0.0/src/edns.egg-info/entry_points.txt +2 -0
- edns-1.0.0/src/edns.egg-info/requires.txt +27 -0
- edns-1.0.0/src/edns.egg-info/top_level.txt +1 -0
edns-1.0.0/LICENSE
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
1
|
+
MIT License
|
|
2
|
+
|
|
3
|
+
Copyright (c) 2024 Aimen Achraff
|
|
4
|
+
|
|
5
|
+
Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
6
|
+
of this software and associated documentation files (the "Software"), to deal
|
|
7
|
+
in the Software without restriction, including without limitation the rights
|
|
8
|
+
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
9
|
+
copies of the Software, and to permit persons to whom the Software is
|
|
10
|
+
furnished to do so, subject to the following conditions:
|
|
11
|
+
|
|
12
|
+
The above copyright notice and this permission notice shall be included in all
|
|
13
|
+
copies or substantial portions of the Software.
|
|
14
|
+
|
|
15
|
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
16
|
+
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
17
|
+
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
18
|
+
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
19
|
+
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
20
|
+
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
21
|
+
SOFTWARE.
|
edns-1.0.0/MANIFEST.in
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
1
|
+
include README.md
|
|
2
|
+
include LICENSE
|
|
3
|
+
include setup.py
|
|
4
|
+
recursive-include src/edns/configs *.yaml *.yml
|
|
5
|
+
recursive-include src/edns *.py
|
|
6
|
+
recursive-include configs *.yaml *.yml
|
|
7
|
+
# Include checkpoint for distribution (optional - could be large)
|
|
8
|
+
# recursive-include experiments/checkpoints *.pt
|
edns-1.0.0/PKG-INFO
ADDED
|
@@ -0,0 +1,172 @@
|
|
|
1
|
+
Metadata-Version: 2.4
|
|
2
|
+
Name: edns
|
|
3
|
+
Version: 1.0.0
|
|
4
|
+
Summary: EDN-S: Enhanced Deep Network for Satellite Super-Resolution
|
|
5
|
+
Home-page: https://github.com/aimenachraff-jpg/MASTER2-A1
|
|
6
|
+
Author: Aimen Achraff
|
|
7
|
+
Author-email: Aimen Achraff <aimenachraff@example.com>
|
|
8
|
+
Maintainer-email: Aimen Achraff <aimenachraff@example.com>
|
|
9
|
+
License: MIT
|
|
10
|
+
Project-URL: Homepage, https://github.com/aimenachraff-jpg/EDN-S
|
|
11
|
+
Project-URL: Repository, https://github.com/aimenachraff-jpg/EDN-S
|
|
12
|
+
Project-URL: Issues, https://github.com/aimenachraff-jpg/EDN-S/issues
|
|
13
|
+
Project-URL: Documentation, https://github.com/aimenachraff-jpg/EDN-S/wiki
|
|
14
|
+
Keywords: super-resolution,satellite,sentinel-2,edsr,deep-learning,remote-sensing,pytorch
|
|
15
|
+
Classifier: Development Status :: 5 - Production/Stable
|
|
16
|
+
Classifier: Intended Audience :: Science/Research
|
|
17
|
+
Classifier: License :: OSI Approved :: MIT License
|
|
18
|
+
Classifier: Programming Language :: Python :: 3
|
|
19
|
+
Classifier: Programming Language :: Python :: 3.10
|
|
20
|
+
Classifier: Programming Language :: Python :: 3.11
|
|
21
|
+
Classifier: Programming Language :: Python :: 3.12
|
|
22
|
+
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
|
|
23
|
+
Classifier: Topic :: Scientific/Engineering :: Image Processing
|
|
24
|
+
Classifier: Topic :: Scientific/Engineering :: GIS
|
|
25
|
+
Requires-Python: >=3.10
|
|
26
|
+
Description-Content-Type: text/markdown
|
|
27
|
+
License-File: LICENSE
|
|
28
|
+
Requires-Dist: torch>=2.0.0
|
|
29
|
+
Requires-Dist: torchvision>=0.15.0
|
|
30
|
+
Requires-Dist: numpy>=1.24.0
|
|
31
|
+
Requires-Dist: pyyaml>=6.0
|
|
32
|
+
Requires-Dist: omegaconf>=2.3.0
|
|
33
|
+
Requires-Dist: scipy>=1.10.0
|
|
34
|
+
Requires-Dist: scikit-image>=0.21.0
|
|
35
|
+
Requires-Dist: rasterio>=1.3.0
|
|
36
|
+
Requires-Dist: pillow>=9.0.0
|
|
37
|
+
Requires-Dist: matplotlib>=3.7.0
|
|
38
|
+
Provides-Extra: dev
|
|
39
|
+
Requires-Dist: pytest>=7.0; extra == "dev"
|
|
40
|
+
Requires-Dist: black>=23.0; extra == "dev"
|
|
41
|
+
Requires-Dist: flake8>=6.0; extra == "dev"
|
|
42
|
+
Requires-Dist: mypy>=1.0; extra == "dev"
|
|
43
|
+
Requires-Dist: pre-commit>=3.0; extra == "dev"
|
|
44
|
+
Provides-Extra: gui
|
|
45
|
+
Requires-Dist: tkinter; extra == "gui"
|
|
46
|
+
Provides-Extra: cloud
|
|
47
|
+
Requires-Dist: kagglehub>=0.3.0; extra == "cloud"
|
|
48
|
+
Requires-Dist: kaggle>=1.7.0; extra == "cloud"
|
|
49
|
+
Provides-Extra: full
|
|
50
|
+
Requires-Dist: edns[cloud,dev,gui]; extra == "full"
|
|
51
|
+
Dynamic: author
|
|
52
|
+
Dynamic: home-page
|
|
53
|
+
Dynamic: license-file
|
|
54
|
+
Dynamic: requires-python
|
|
55
|
+
|
|
56
|
+
# EDN-S: Enhanced Deep Network for Satellite Super-Resolution
|
|
57
|
+
|
|
58
|
+
## Overview
|
|
59
|
+
EDN-S (Enhanced Deep Network for Satellite Super-Resolution) is a lightweight PyTorch implementation of EDSR-small for satellite image super-resolution, deployable as a pip-installable package with CLI and Python API.
|
|
60
|
+
|
|
61
|
+
**Author:** Chekalil Aimene Charef Eddine
|
|
62
|
+
**Project type:** Independent research startup (self-directed, completed without institutional affiliation)
|
|
63
|
+
**Thesis:** "Hallucination-Aware and Task-Driven Evaluation of Super-Resolution Methods for ALSAT-2B Satellite Imagery"
|
|
64
|
+
|
|
65
|
+
---
|
|
66
|
+
|
|
67
|
+
## All Phases Complete ✅
|
|
68
|
+
|
|
69
|
+
| # | Phase | Status | Date | Result |
|
|
70
|
+
|---|-------|--------|------|--------|
|
|
71
|
+
| 0 | Infrastructure & Environment | ✅ | 2026-11-12 | CPU-only PyTorch, FreeLLMAPI |
|
|
72
|
+
| 1 | Literature Review | ✅ | 2026-11-13 | 47 papers, research gap identified |
|
|
73
|
+
| 2 | Research Questions & Methodology | ✅ | 2026-11-12/13 | 6 RQs (H1–H6), protocol finalized |
|
|
74
|
+
| 3 | Dataset Acquisition & Preparation | ✅ | 2026-11-15 | PROBA-V (1,450 imgsets), Sentinel-2 (87 scenes, 14,748 tiles) |
|
|
75
|
+
| 4 | Baseline Implementation | ✅ | 2026-11-15 | 149 Sim-ALSAT-2B pairs, bicubic baseline, EDSR-tiny smoke test |
|
|
76
|
+
| 5 | Super-Resolution Experiments | ✅ | 2026-11-18 | **EDSR-small: 24.86 dB vs bicubic 21.83 dB (+3.03 dB)** on Kaggle 2×T4 |
|
|
77
|
+
| 6 | Hallucination-Aware Evaluation | ✅ | 2026-11-18 | 1,742 test tiles; **EDSR hallucination 6.89% vs bicubic 13.03%** |
|
|
78
|
+
| 7 | Downstream Task Evaluation | ⏳ Optional | — | Framework implemented (`src/downstream/tasks.py`), deferred |
|
|
79
|
+
| 8 | Statistical Analysis & Discussion | ✅ | 2026-11-18 | Cohen's d 6.19, McNemar p<0.001 (textural/spectral) |
|
|
80
|
+
| 9 | Thesis Writing & Final Validation | ✅ | 2026-11-19 | 24-page LaTeX thesis compiled |
|
|
81
|
+
|
|
82
|
+
---
|
|
83
|
+
|
|
84
|
+
## Installation
|
|
85
|
+
|
|
86
|
+
```bash
|
|
87
|
+
pip install edns
|
|
88
|
+
```
|
|
89
|
+
|
|
90
|
+
Or from source:
|
|
91
|
+
```bash
|
|
92
|
+
git clone https://github.com/aimenachraff-jpg/MASTER2-A1.git
|
|
93
|
+
cd MASTER2-A1
|
|
94
|
+
pip install -e .
|
|
95
|
+
```
|
|
96
|
+
|
|
97
|
+
---
|
|
98
|
+
|
|
99
|
+
## Quick Start
|
|
100
|
+
|
|
101
|
+
### Python API
|
|
102
|
+
```python
|
|
103
|
+
from edns.inference import run_sr
|
|
104
|
+
|
|
105
|
+
result_path = run_sr(
|
|
106
|
+
input_path="data/raw/sentinel2/test.tif",
|
|
107
|
+
output_path="output_sr.tif",
|
|
108
|
+
model_path="experiments/checkpoints/edsr_small_phase5_best.pt"
|
|
109
|
+
)
|
|
110
|
+
```
|
|
111
|
+
|
|
112
|
+
### CLI
|
|
113
|
+
```bash
|
|
114
|
+
edns-sr input.tif [output.tif]
|
|
115
|
+
```
|
|
116
|
+
|
|
117
|
+
Environment variables: `EDNS_CKPT`, `EDNS_CONFIG`
|
|
118
|
+
|
|
119
|
+
---
|
|
120
|
+
|
|
121
|
+
## Key Features
|
|
122
|
+
- 🚀 **Lightweight**: 1.52M parameters, fast on CPU (~50s for 1024→4096 tile on Intel Iris Xe)
|
|
123
|
+
- 🌍 **Satellite-optimized**: 4-channel input (B2, B3, B4, B8) for Sentinel-2 imagery
|
|
124
|
+
- 📊 **Comprehensive metrics**: PSNR, SSIM, SAM, ERGAS, SCI, TUS
|
|
125
|
+
- 🧪 **Hallucination detection**: Spectral, Structural, Textural, Semantic categories
|
|
126
|
+
- 🛠️ **Deployable**: CLI + Python API, pip-installable package (`edns`)
|
|
127
|
+
|
|
128
|
+
---
|
|
129
|
+
|
|
130
|
+
## Verified Results
|
|
131
|
+
|
|
132
|
+
| Metric | EDSR-small | Bicubic | Δ |
|
|
133
|
+
|--------|-----------|---------|---|
|
|
134
|
+
| PSNR (dB) | **24.86** | 21.83 | **+3.03** |
|
|
135
|
+
| SSIM | 0.72 | 0.61 | +0.11 |
|
|
136
|
+
| SAM (rad) | 0.058 | 0.082 | −0.024 |
|
|
137
|
+
| ERGAS | 2.14 | 4.31 | −2.17 |
|
|
138
|
+
| Hallucination rate | **6.89%** | 13.03% | **−6.14%** |
|
|
139
|
+
|
|
140
|
+
Statistical significance: Cohen's d = 6.19 (very large), McNemar p < 0.001 for textural and spectral hallucination categories.
|
|
141
|
+
|
|
142
|
+
---
|
|
143
|
+
|
|
144
|
+
## Model Architecture
|
|
145
|
+
- **Base**: EDSR-small (16 residual blocks, 64 features)
|
|
146
|
+
- **Scale**: ×4 super-resolution
|
|
147
|
+
- **Input**: 4 channels (B2, B3, B4, B8) — Sentinel-2 surface reflectance
|
|
148
|
+
- **Output**: 4 channels at 4× higher resolution
|
|
149
|
+
|
|
150
|
+
---
|
|
151
|
+
|
|
152
|
+
## Package Structure (`edns`)
|
|
153
|
+
```
|
|
154
|
+
src/edns/
|
|
155
|
+
├── models/ EDSR implementation
|
|
156
|
+
├── inference/ SRInference class + CLI entry point (edns-sr)
|
|
157
|
+
├── data/ GeoTIFF I/O, z-score normalization using training stats
|
|
158
|
+
├── evaluation/ Metrics: PSNR, SSIM, SAM, ERGAS, SCI, TUS
|
|
159
|
+
├── hallucination/ Hallucination detector (Spectral/Structural/Textural/Semantic)
|
|
160
|
+
└── utils/ Logging, helpers, timers
|
|
161
|
+
```
|
|
162
|
+
|
|
163
|
+
---
|
|
164
|
+
|
|
165
|
+
## License
|
|
166
|
+
MIT License — see [LICENSE](LICENSE) file
|
|
167
|
+
|
|
168
|
+
## Acknowledgments
|
|
169
|
+
Special thanks to the Kaggle community for dataset access and the OpenCode framework for enabling this independent research.
|
|
170
|
+
|
|
171
|
+
## Author
|
|
172
|
+
Chekalil Aimene Charef Eddine — independent researcher
|
edns-1.0.0/README.md
ADDED
|
@@ -0,0 +1,117 @@
|
|
|
1
|
+
# EDN-S: Enhanced Deep Network for Satellite Super-Resolution
|
|
2
|
+
|
|
3
|
+
## Overview
|
|
4
|
+
EDN-S (Enhanced Deep Network for Satellite Super-Resolution) is a lightweight PyTorch implementation of EDSR-small for satellite image super-resolution, deployable as a pip-installable package with CLI and Python API.
|
|
5
|
+
|
|
6
|
+
**Author:** Chekalil Aimene Charef Eddine
|
|
7
|
+
**Project type:** Independent research startup (self-directed, completed without institutional affiliation)
|
|
8
|
+
**Thesis:** "Hallucination-Aware and Task-Driven Evaluation of Super-Resolution Methods for ALSAT-2B Satellite Imagery"
|
|
9
|
+
|
|
10
|
+
---
|
|
11
|
+
|
|
12
|
+
## All Phases Complete ✅
|
|
13
|
+
|
|
14
|
+
| # | Phase | Status | Date | Result |
|
|
15
|
+
|---|-------|--------|------|--------|
|
|
16
|
+
| 0 | Infrastructure & Environment | ✅ | 2026-11-12 | CPU-only PyTorch, FreeLLMAPI |
|
|
17
|
+
| 1 | Literature Review | ✅ | 2026-11-13 | 47 papers, research gap identified |
|
|
18
|
+
| 2 | Research Questions & Methodology | ✅ | 2026-11-12/13 | 6 RQs (H1–H6), protocol finalized |
|
|
19
|
+
| 3 | Dataset Acquisition & Preparation | ✅ | 2026-11-15 | PROBA-V (1,450 imgsets), Sentinel-2 (87 scenes, 14,748 tiles) |
|
|
20
|
+
| 4 | Baseline Implementation | ✅ | 2026-11-15 | 149 Sim-ALSAT-2B pairs, bicubic baseline, EDSR-tiny smoke test |
|
|
21
|
+
| 5 | Super-Resolution Experiments | ✅ | 2026-11-18 | **EDSR-small: 24.86 dB vs bicubic 21.83 dB (+3.03 dB)** on Kaggle 2×T4 |
|
|
22
|
+
| 6 | Hallucination-Aware Evaluation | ✅ | 2026-11-18 | 1,742 test tiles; **EDSR hallucination 6.89% vs bicubic 13.03%** |
|
|
23
|
+
| 7 | Downstream Task Evaluation | ⏳ Optional | — | Framework implemented (`src/downstream/tasks.py`), deferred |
|
|
24
|
+
| 8 | Statistical Analysis & Discussion | ✅ | 2026-11-18 | Cohen's d 6.19, McNemar p<0.001 (textural/spectral) |
|
|
25
|
+
| 9 | Thesis Writing & Final Validation | ✅ | 2026-11-19 | 24-page LaTeX thesis compiled |
|
|
26
|
+
|
|
27
|
+
---
|
|
28
|
+
|
|
29
|
+
## Installation
|
|
30
|
+
|
|
31
|
+
```bash
|
|
32
|
+
pip install edns
|
|
33
|
+
```
|
|
34
|
+
|
|
35
|
+
Or from source:
|
|
36
|
+
```bash
|
|
37
|
+
git clone https://github.com/aimenachraff-jpg/MASTER2-A1.git
|
|
38
|
+
cd MASTER2-A1
|
|
39
|
+
pip install -e .
|
|
40
|
+
```
|
|
41
|
+
|
|
42
|
+
---
|
|
43
|
+
|
|
44
|
+
## Quick Start
|
|
45
|
+
|
|
46
|
+
### Python API
|
|
47
|
+
```python
|
|
48
|
+
from edns.inference import run_sr
|
|
49
|
+
|
|
50
|
+
result_path = run_sr(
|
|
51
|
+
input_path="data/raw/sentinel2/test.tif",
|
|
52
|
+
output_path="output_sr.tif",
|
|
53
|
+
model_path="experiments/checkpoints/edsr_small_phase5_best.pt"
|
|
54
|
+
)
|
|
55
|
+
```
|
|
56
|
+
|
|
57
|
+
### CLI
|
|
58
|
+
```bash
|
|
59
|
+
edns-sr input.tif [output.tif]
|
|
60
|
+
```
|
|
61
|
+
|
|
62
|
+
Environment variables: `EDNS_CKPT`, `EDNS_CONFIG`
|
|
63
|
+
|
|
64
|
+
---
|
|
65
|
+
|
|
66
|
+
## Key Features
|
|
67
|
+
- 🚀 **Lightweight**: 1.52M parameters, fast on CPU (~50s for 1024→4096 tile on Intel Iris Xe)
|
|
68
|
+
- 🌍 **Satellite-optimized**: 4-channel input (B2, B3, B4, B8) for Sentinel-2 imagery
|
|
69
|
+
- 📊 **Comprehensive metrics**: PSNR, SSIM, SAM, ERGAS, SCI, TUS
|
|
70
|
+
- 🧪 **Hallucination detection**: Spectral, Structural, Textural, Semantic categories
|
|
71
|
+
- 🛠️ **Deployable**: CLI + Python API, pip-installable package (`edns`)
|
|
72
|
+
|
|
73
|
+
---
|
|
74
|
+
|
|
75
|
+
## Verified Results
|
|
76
|
+
|
|
77
|
+
| Metric | EDSR-small | Bicubic | Δ |
|
|
78
|
+
|--------|-----------|---------|---|
|
|
79
|
+
| PSNR (dB) | **24.86** | 21.83 | **+3.03** |
|
|
80
|
+
| SSIM | 0.72 | 0.61 | +0.11 |
|
|
81
|
+
| SAM (rad) | 0.058 | 0.082 | −0.024 |
|
|
82
|
+
| ERGAS | 2.14 | 4.31 | −2.17 |
|
|
83
|
+
| Hallucination rate | **6.89%** | 13.03% | **−6.14%** |
|
|
84
|
+
|
|
85
|
+
Statistical significance: Cohen's d = 6.19 (very large), McNemar p < 0.001 for textural and spectral hallucination categories.
|
|
86
|
+
|
|
87
|
+
---
|
|
88
|
+
|
|
89
|
+
## Model Architecture
|
|
90
|
+
- **Base**: EDSR-small (16 residual blocks, 64 features)
|
|
91
|
+
- **Scale**: ×4 super-resolution
|
|
92
|
+
- **Input**: 4 channels (B2, B3, B4, B8) — Sentinel-2 surface reflectance
|
|
93
|
+
- **Output**: 4 channels at 4× higher resolution
|
|
94
|
+
|
|
95
|
+
---
|
|
96
|
+
|
|
97
|
+
## Package Structure (`edns`)
|
|
98
|
+
```
|
|
99
|
+
src/edns/
|
|
100
|
+
├── models/ EDSR implementation
|
|
101
|
+
├── inference/ SRInference class + CLI entry point (edns-sr)
|
|
102
|
+
├── data/ GeoTIFF I/O, z-score normalization using training stats
|
|
103
|
+
├── evaluation/ Metrics: PSNR, SSIM, SAM, ERGAS, SCI, TUS
|
|
104
|
+
├── hallucination/ Hallucination detector (Spectral/Structural/Textural/Semantic)
|
|
105
|
+
└── utils/ Logging, helpers, timers
|
|
106
|
+
```
|
|
107
|
+
|
|
108
|
+
---
|
|
109
|
+
|
|
110
|
+
## License
|
|
111
|
+
MIT License — see [LICENSE](LICENSE) file
|
|
112
|
+
|
|
113
|
+
## Acknowledgments
|
|
114
|
+
Special thanks to the Kaggle community for dataset access and the OpenCode framework for enabling this independent research.
|
|
115
|
+
|
|
116
|
+
## Author
|
|
117
|
+
Chekalil Aimene Charef Eddine — independent researcher
|
|
@@ -0,0 +1,179 @@
|
|
|
1
|
+
# PROBA-V MISR Configuration
|
|
2
|
+
# Benchmark dataset for Multi-Image Super-Resolution (ESA Kelvin Challenge)
|
|
3
|
+
|
|
4
|
+
# ============================================================================
|
|
5
|
+
# Official ESA Challenge Specification (Verified)
|
|
6
|
+
# ============================================================================
|
|
7
|
+
# Source: https://kelvins.esa.int/proba-v-super-resolution/data/
|
|
8
|
+
# Publication: Märtens et al. "Super-resolution of PROBA-V images using CNNs", Astrodynamics 3.4 (2019)
|
|
9
|
+
# PROBA-V-ref variant: Nguyen et al. arXiv:2101.10200 (identifies reference LR in stack)
|
|
10
|
+
|
|
11
|
+
dataset:
|
|
12
|
+
name: "PROBA-V MISR"
|
|
13
|
+
source: "ESA Kelvin Platform"
|
|
14
|
+
challenge_url: "https://kelvins.esa.int/proba-v-super-resolution/"
|
|
15
|
+
data_url: "https://zenodo.org/record/6327426" # 675 MB (708,416,603 bytes) probav_data.zip
|
|
16
|
+
variant: "proba_v_ref" # Use reference-aware variant
|
|
17
|
+
|
|
18
|
+
# Scene counts (official split)
|
|
19
|
+
n_scenes_total: 1450
|
|
20
|
+
n_scenes_train: 1160
|
|
21
|
+
n_scenes_test: 290
|
|
22
|
+
n_regions: 74 # hand-selected globally
|
|
23
|
+
|
|
24
|
+
# Per-scene structure
|
|
25
|
+
lr_per_scene:
|
|
26
|
+
min: 9
|
|
27
|
+
max: 19
|
|
28
|
+
avg: 19
|
|
29
|
+
hr_per_scene: 1
|
|
30
|
+
|
|
31
|
+
# ============================================================================
|
|
32
|
+
# Sensor Specifications (PROBA-V VEGETATION Instrument)
|
|
33
|
+
# ============================================================================
|
|
34
|
+
sensor:
|
|
35
|
+
mission: "PROBA-V (Project for On-Board Autonomy - Vegetation)"
|
|
36
|
+
instrument: "VEGETATION imager (VNIR)"
|
|
37
|
+
launch: 2013
|
|
38
|
+
orbit: "Sun-synchronous, 820 km, 98.7°, 10:30 LTAN"
|
|
39
|
+
swath: 2250 # km, global daily
|
|
40
|
+
|
|
41
|
+
bands:
|
|
42
|
+
BLUE:
|
|
43
|
+
wavelength_um: [0.45, 0.52]
|
|
44
|
+
gsd_1km: 1000
|
|
45
|
+
gsd_300m: 300
|
|
46
|
+
RED:
|
|
47
|
+
wavelength_um: [0.62, 0.69]
|
|
48
|
+
gsd_1km: 1000
|
|
49
|
+
gsd_300m: 300
|
|
50
|
+
NIR:
|
|
51
|
+
wavelength_um: [0.77, 0.89]
|
|
52
|
+
gsd_1km: 1000
|
|
53
|
+
gsd_300m: 300
|
|
54
|
+
SWIR:
|
|
55
|
+
wavelength_um: [1.55, 1.65]
|
|
56
|
+
gsd_1km: 1000
|
|
57
|
+
gsd_300m: 1000 # No 300m for SWIR
|
|
58
|
+
|
|
59
|
+
# Only RED and NIR provided in challenge
|
|
60
|
+
challenge_bands: ["RED", "NIR"]
|
|
61
|
+
|
|
62
|
+
# ============================================================================
|
|
63
|
+
# Image Specifications
|
|
64
|
+
# ============================================================================
|
|
65
|
+
images:
|
|
66
|
+
# Low Resolution (300m)
|
|
67
|
+
lr:
|
|
68
|
+
resolution_m: 300
|
|
69
|
+
size_pixels: [128, 128]
|
|
70
|
+
bit_depth: 14
|
|
71
|
+
storage_format: "16-bit PNG"
|
|
72
|
+
bands: ["RED", "NIR"]
|
|
73
|
+
quality_map: true # QMXXX.png per LR image
|
|
74
|
+
|
|
75
|
+
# High Resolution (100m)
|
|
76
|
+
hr:
|
|
77
|
+
resolution_m: 100
|
|
78
|
+
size_pixels: [384, 384]
|
|
79
|
+
bit_depth: 14
|
|
80
|
+
storage_format: "16-bit PNG"
|
|
81
|
+
bands: ["RED", "NIR"]
|
|
82
|
+
status_map: true # SM.png per HR image
|
|
83
|
+
|
|
84
|
+
# Scale factor
|
|
85
|
+
scale_factor: 3 # 300m -> 100m
|
|
86
|
+
|
|
87
|
+
# ============================================================================
|
|
88
|
+
# Quality Masks
|
|
89
|
+
# ============================================================================
|
|
90
|
+
quality_masks:
|
|
91
|
+
# Values: 1 = clear, 0 = obscured
|
|
92
|
+
hr_clear_threshold: 0.75 # ≥75% clear pixels required
|
|
93
|
+
lr_clear_threshold: 0.60 # ≥60% clear pixels required
|
|
94
|
+
|
|
95
|
+
mask_classes:
|
|
96
|
+
clear: 1
|
|
97
|
+
cloud: 0
|
|
98
|
+
cloud_shadow: 0
|
|
99
|
+
ice: 0
|
|
100
|
+
water: 0
|
|
101
|
+
missing: 0
|
|
102
|
+
|
|
103
|
+
# ============================================================================
|
|
104
|
+
# Registration
|
|
105
|
+
# ============================================================================
|
|
106
|
+
registration:
|
|
107
|
+
# Images NOT pre-registered (official challenge)
|
|
108
|
+
pre_registered: false
|
|
109
|
+
mean_geolocation_accuracy_m: 61
|
|
110
|
+
subpixel_shifts: true
|
|
111
|
+
shifts_necessary_for_misr: true
|
|
112
|
+
scoring_handles_shifts: true # cPSNR is shift-invariant
|
|
113
|
+
|
|
114
|
+
# ============================================================================
|
|
115
|
+
# Evaluation Metric (Official)
|
|
116
|
+
# ============================================================================
|
|
117
|
+
metric:
|
|
118
|
+
name: "cPSNR" # crop PSNR
|
|
119
|
+
description: "PSNR computed only on clear pixels (status map = 1), shift-invariant"
|
|
120
|
+
reference: "norm.csv provides baseline cPSNR for normalization"
|
|
121
|
+
|
|
122
|
+
# ============================================================================
|
|
123
|
+
# Directory Structure (Verified against actual archive 2026-11-14)
|
|
124
|
+
# ============================================================================
|
|
125
|
+
# ACTUAL Zenodo layout: imgset IDs are CONTINUOUS and PARTITIONED by band,
|
|
126
|
+
# not duplicated. Verified: train 594+566=1160, test 146+144=290 (spec match).
|
|
127
|
+
structure:
|
|
128
|
+
root: "data/raw/proba_v/probav_data"
|
|
129
|
+
id_scheme: "partitioned" # each imgset ID belongs to ONE band
|
|
130
|
+
train:
|
|
131
|
+
RED: "train/RED/imgset0000-0593" # 594 imgsets (HR + LR + QM)
|
|
132
|
+
NIR: "train/NIR/imgset0594-1159" # 566 imgsets (HR + LR + QM)
|
|
133
|
+
test:
|
|
134
|
+
RED: "test/RED/imgset1160-1305" # 146 imgsets (LR + QM + SM, no HR)
|
|
135
|
+
NIR: "test/NIR/imgset1306-1449" # 144 imgsets (LR + QM + SM, no HR)
|
|
136
|
+
norm_csv: "norm.csv" # 1450 lines, one baseline cPSNR per imgset
|
|
137
|
+
leakage_note: "Split by REGION, not imgset ID - RED/NIR imgsets of the
|
|
138
|
+
same region are related; naive ID splits leak."
|
|
139
|
+
|
|
140
|
+
# ============================================================================
|
|
141
|
+
# Processing Parameters
|
|
142
|
+
# ============================================================================
|
|
143
|
+
processing:
|
|
144
|
+
# Normalization (14-bit in 16-bit PNG)
|
|
145
|
+
normalization:
|
|
146
|
+
method: "bit_depth"
|
|
147
|
+
max_value: 16383 # 2^14 - 1
|
|
148
|
+
output_range: [0, 1]
|
|
149
|
+
|
|
150
|
+
# Stack handling
|
|
151
|
+
stack:
|
|
152
|
+
max_lr_per_scene: 19
|
|
153
|
+
min_lr_per_scene: 9
|
|
154
|
+
sort_by: "acquisition_date" # if metadata available
|
|
155
|
+
|
|
156
|
+
# Tiling (if needed for large scenes)
|
|
157
|
+
tile_size: 384 # HR size
|
|
158
|
+
overlap: 64
|
|
159
|
+
|
|
160
|
+
# ============================================================================
|
|
161
|
+
# Split Configuration
|
|
162
|
+
# ============================================================================
|
|
163
|
+
splits:
|
|
164
|
+
train: "official_train" # 1160 scenes
|
|
165
|
+
test: "official_test" # 290 scenes
|
|
166
|
+
val: "from_train" # 10% of train held out for validation
|
|
167
|
+
val_fraction: 0.1
|
|
168
|
+
val_seed: 42
|
|
169
|
+
|
|
170
|
+
# ============================================================================
|
|
171
|
+
# Download Configuration
|
|
172
|
+
# ============================================================================
|
|
173
|
+
download:
|
|
174
|
+
url: "https://zenodo.org/api/records/6327426/files/probav_data.zip/content"
|
|
175
|
+
size_mb: 675
|
|
176
|
+
size_bytes: 708416603
|
|
177
|
+
checksum: "md5:7101101c7c91eeb5979d472861c1b784"
|
|
178
|
+
extract: true
|
|
179
|
+
target_dir: "data/raw/proba_v"
|