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.
Files changed (36) hide show
  1. edns-1.0.0/LICENSE +21 -0
  2. edns-1.0.0/MANIFEST.in +8 -0
  3. edns-1.0.0/PKG-INFO +172 -0
  4. edns-1.0.0/README.md +117 -0
  5. edns-1.0.0/configs/data/probav.yaml +179 -0
  6. edns-1.0.0/configs/data/sentinel2.yaml +220 -0
  7. edns-1.0.0/configs/degradation/alsat2b.yaml +137 -0
  8. edns-1.0.0/configs/models/edsr_small.yaml +30 -0
  9. edns-1.0.0/configs/models/edsr_tiny.yaml +28 -0
  10. edns-1.0.0/configs/models/realesrgan.yaml +50 -0
  11. edns-1.0.0/configs/models/swinir_base.yaml +39 -0
  12. edns-1.0.0/pyproject.toml +110 -0
  13. edns-1.0.0/setup.cfg +4 -0
  14. edns-1.0.0/setup.py +62 -0
  15. edns-1.0.0/src/edns/__init__.py +60 -0
  16. edns-1.0.0/src/edns/configs/models/edsr_small.yaml +30 -0
  17. edns-1.0.0/src/edns/data/__init__.py +33 -0
  18. edns-1.0.0/src/edns/data/io.py +124 -0
  19. edns-1.0.0/src/edns/data/normalization.py +190 -0
  20. edns-1.0.0/src/edns/evaluation/__init__.py +22 -0
  21. edns-1.0.0/src/edns/evaluation/metrics.py +432 -0
  22. edns-1.0.0/src/edns/hallucination/__init__.py +31 -0
  23. edns-1.0.0/src/edns/hallucination/detection.py +732 -0
  24. edns-1.0.0/src/edns/inference/__init__.py +14 -0
  25. edns-1.0.0/src/edns/inference/app.py +335 -0
  26. edns-1.0.0/src/edns/models/__init__.py +13 -0
  27. edns-1.0.0/src/edns/models/edsr.py +218 -0
  28. edns-1.0.0/src/edns/utils/__init__.py +15 -0
  29. edns-1.0.0/src/edns/utils/helpers.py +90 -0
  30. edns-1.0.0/src/edns/utils/logging.py +53 -0
  31. edns-1.0.0/src/edns.egg-info/PKG-INFO +172 -0
  32. edns-1.0.0/src/edns.egg-info/SOURCES.txt +34 -0
  33. edns-1.0.0/src/edns.egg-info/dependency_links.txt +1 -0
  34. edns-1.0.0/src/edns.egg-info/entry_points.txt +2 -0
  35. edns-1.0.0/src/edns.egg-info/requires.txt +27 -0
  36. 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"