Revise dataset README: 14-class only, add spatial/data tables and figure section; remove Croissant section

README.md

@@ -12,8 +12,6 @@ tags:
 - geospatial
 - drone-imagery
 - semantic-segmentation
-size_categories:
-- n<1K
 ---
 
 # GoldMiningMDD
@@ -21,23 +19,8 @@ size_categories:
 GoldMiningMDD is a drone orthomosaic semantic-segmentation dataset for mining-site analysis in Madre de Dios, Peru.
 
 - 12 sites
-- 24 PNG files total (12 images + 12 labels)
-- Splits: train/val/test at site level
-- Labels: 14 foreground classes (`1..14`) + background (`0`)
-
-## Directory structure
-
-```text
-train/
-  image/*.png
-  label/*.png
-val/
-  image/*.png
-  label/*.png
-test/
-  image/*.png
-  label/*.png
-```
+- Site-level split: train / val / test
+- Image-label pairs are aligned and share the same spatial footprint per site
 
 ## Split definition
 
@@ -49,12 +32,10 @@ test/
 - `PlayaMirador1`: bottom half (train)
 - `PlayaMirador2`: top half (val)
 
-## Label schema (merged classes)
-
-Class IDs are globally consistent across all sites.
+## Class definition (14 classes + background)
 
 | ID | Class |
-|---:|---|
+| ---: | --- |
 | 0 | Background |
 | 1 | Building |
 | 2 | Mining raft |
@@ -71,46 +52,53 @@ Class IDs are globally consistent across all sites.
 | 13 | Sluice |
 | 14 | Vehicles |
 
-### Merge rules from original 19-class labels
-
-- `Building` = original IDs `{1 (Urban area), 4 (Mining camp)}`
-- `Heavy machinery` = `{5 (Front loader), 8 (Excavator), 9 (Heavy machinery), 19 (Dump truck)}`
-- `Vehicles` = `{17 (Vehicles), 18 (Small vehicles)}`
-- Other classes map 1-to-1 to their corresponding semantic category.
-
-## Image-label pairing
-
-- Pairing is by basename under each split:
-  - `train/image/AcumulacionAaron2B.png` ↔ `train/label/AcumulacionAaron2B.png`
-- Label rasters are aligned to image spatial extent and grid before export.
-
-## Spatial/reference notes
-
-- Source CRS in preprocessing: EPSG:32719
-- This repository stores aligned PNG exports for training/evaluation.
-- Full spatial metadata and preprocessing details are maintained in the project-side documentation.
-
-## Recommended usage
-
-- Treat label value `0` as background/ignore if your training protocol does not score background.
-- Keep site-level split boundaries (no cross-split mixing of sites).
-
-## Croissant status
-
-Endpoint:
-
-```text
-https://huggingface.co/api/datasets/kangnicui2/GoldMiningMDD/croissant
-```
-
-Current behavior for this repository: the endpoint is not available yet.
-
-Important platform note: the Hugging Face Dataset Viewer/Croissant service is powered by `datasets-server`, and private datasets require a plan that supports private dataset-server features (for many accounts this means PRO/Enterprise).
-
-How to get Croissant for this dataset:
 
-1. Keep this repo as dataset format compatible with Parquet conversion (already fine for Hub processing).
-2. Make sure dataset-server can run on this repo:
-   - Option A: make the dataset public; or
-   - Option B: use an account/org plan that supports private dataset-server.
-3. Once processed, query the endpoint above and use the returned JSON-LD as machine-readable Croissant metadata.
+## Spatial metadata table (source files + aligned output)
+
+| Site | Ortho file | Label source file | CRS | Ortho size (W x H px) | Label size (W x H px) | Ortho res (m/px x,y) | Label res (m/px x,y) | Ortho lon range | Ortho lat range | Label lon range | Label lat range | Output size (W x H px) | IoU | Acquisition date | Split |
+| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ---: | --- | --- |
+| AcumulacionAaron2B | `AcumulacionAaron2B_corrected.tif` | `AcumulacionAaron2B_classified_corrected.tif` | EPSG:32719 | 38122x17643 | 38111x17760 | 0.056308, 0.056308 | 0.056323, 0.055936 | [-70.419597, -70.399788] | [-12.627643, -12.618558] | [-70.419597, -70.399788] | [-12.627643, -12.618558] | 38122x17643 | 1.0000 | 2022-04-24 | train |
+| Anel | `Anel_corrected.tif` | `Anel_classified_corrected.tif` | EPSG:32719 | 26909x24725 | 17662x17661 | 0.056620, 0.056620 | 0.058580, 0.059049 | [-69.714044, -69.699976] | [-12.713767, -12.701071] | [-69.711529, -69.701974] | [-12.711964, -12.702509] | 18274x18420 | 0.5059 | 2022-04-08 | test |
+| Clavelito | `Clavelito_corrected.tif` | `Clavelito_classified_corrected.tif` | EPSG:32719 | 19865x40307 | 19357x39684 | 0.075601, 0.075601 | 0.076376, 0.076781 | [-70.605945, -70.591929] | [-12.961391, -12.933761] | [-70.605944, -70.592143] | [-12.961387, -12.933761] | 19556x40304 | 0.9843 | 2022-05-09 | val |
+| ElEngano | `ElEngano_corrected.tif` | `ElEngano_classified_corrected.tif` | EPSG:32719 | 23616x24742 | 17473x17473 | 0.057233, 0.057233 | 0.057566, 0.058201 | [-69.962478, -69.949966] | [-13.020161, -13.007312] | [-69.960769, -69.951459] | [-13.018407, -13.009178] | 17575x17770 | 0.5344 | 2022-03-03 | test |
+| Kotsimba | `Kotsimba_corrected.tif` | `Kotsimba_classified_corrected.tif` | EPSG:32719 | 57331x53961 | 57227x53950 | 0.072760, 0.072760 | 0.072889, 0.072771 | [-70.283655, -70.244998] | [-13.136635, -13.100951] | [-70.283654, -70.244999] | [-13.136634, -13.100952] | 57328x53959 | 0.9999 | 2022-05-26 | train |
+| Linda | `Linda2_corrected.tif` | `Linda_classified_corrected.tif` | EPSG:32719 | 23601x23698 | 17352x17352 | 0.057632, 0.057632 | 0.059290, 0.059408 | [-69.953378, -69.940789] | [-13.020082, -13.007688] | [-69.951874, -69.942353] | [-13.018530, -13.009175] | 17852x17888 | 0.5709 | 2022-02-28 | test |
+| Los5Rebeldes | `Los5Rebeldes_corrected.tif` | `Los5Rebeldes_classified_corrected.tif` | EPSG:32719 | 43867x76279 | 33334x66668 | 0.030000, 0.030000 | 0.030140, 0.030001 | [-70.672354, -70.660084] | [-13.034336, -13.013573] | [-70.670932, -70.661549] | [-13.032969, -13.014831] | 33490x66672 | 0.6673 | 2022-02-04 | train |
+| Nayda | `Nayda_corrected.tif` | `Nayda_classified_corrected.tif` | EPSG:32719 | 22677x23480 | 17665x17665 | 0.056610, 0.056610 | 0.057532, 0.057385 | [-69.721603, -69.709746] | [-12.713363, -12.701313] | [-69.720152, -69.710766] | [-12.711798, -12.702607] | 17954x17908 | 0.6038 | 2022-08-04 | val |
+| Paolita | `Paolita1_corrected.tif` | `Paolita1_classified_corrected.tif` | EPSG:32719 | 58663x33523 | 54108x27055 | 0.055446, 0.055446 | 0.055275, 0.055441 | [-69.630457, -69.600462] | [-12.689889, -12.673012] | [-69.629627, -69.602050] | [-12.688983, -12.675355] | 53943x27054 | 0.7420 | 2022-02-21 | test |
+| PlayaMirador1 | `PlayaMirador_corrected.tif` | `PlayaMirador_classified_corrected.tif` | EPSG:32719 | 22911x22394 | 17608x17608 | 0.056792, 0.056792 | 0.056840, 0.057298 | [-70.377536, -70.365474] | [-13.063390, -13.051829] | [-70.376072, -70.366793] | [-13.062030, -13.057446] | 17624x8883 | 0.6102 | 2022-03-10 | train |
+| PlayaMirador2 | `PlayaMirador_corrected.tif` | `PlayaMirador_classified_corrected.tif` | EPSG:32719 | 22911x22394 | 17608x17608 | 0.056792, 0.056792 | 0.056840, 0.057298 | [-70.377536, -70.365474] | [-13.063390, -13.051829] | [-70.376072, -70.366793] | [-13.057446, -13.052861] | 17624x8883 | 0.6102 | 2022-03-10 | val |
+| SantaInesDosMil | `SantaInesDosMil_corrected.tif` | `SantaInesDosMil_classified_corrected.tif` | EPSG:32719 | 22883x24295 | 17185x17184 | 0.058193, 0.058193 | 0.058873, 0.058471 | [-70.386657, -70.374307] | [-13.063436, -13.050589] | [-70.385274, -70.375895] | [-13.061789, -13.052656] | 17387x17267 | 0.5399 | 2022-03-10 | test |
+
+- Note: low IoU means the orthomosaic footprint is larger than the label footprint (not necessarily mismatched).
+- Note: `PlayaMirador` is additionally split in GoldMDD into `PlayaMirador1` (bottom half, train) and `PlayaMirador2` (top half, val). The two rows above use the same source files and split the label latitude range at the midpoint.
+
+## Data location table
+
+| Split | Site | Image path | Label path |
+| --- | --- | --- | --- |
+| train | AcumulacionAaron2B | `train/image/AcumulacionAaron2B.png` | `train/label/AcumulacionAaron2B.png` |
+| train | Kotsimba | `train/image/Kotsimba.png` | `train/label/Kotsimba.png` |
+| train | Los5Rebeldes | `train/image/Los5Rebeldes.png` | `train/label/Los5Rebeldes.png` |
+| train | PlayaMirador1 | `train/image/PlayaMirador1.png` | `train/label/PlayaMirador1.png` |
+| val | Clavelito | `val/image/Clavelito.png` | `val/label/Clavelito.png` |
+| val | Nayda | `val/image/Nayda.png` | `val/label/Nayda.png` |
+| val | PlayaMirador2 | `val/image/PlayaMirador2.png` | `val/label/PlayaMirador2.png` |
+| test | Anel | `test/image/Anel.png` | `test/label/Anel.png` |
+| test | ElEngano | `test/image/ElEngano.png` | `test/label/ElEngano.png` |
+| test | Linda | `test/image/Linda.png` | `test/label/Linda.png` |
+| test | Paolita | `test/image/Paolita.png` | `test/label/Paolita.png` |
+| test | SantaInesDosMil | `test/image/SantaInesDosMil.png` | `test/label/SantaInesDosMil.png` |
+
+## Dataset visualizations
+
+These figures are included in this repository under `figures/`.
+
+### 1) Site spatial distribution
+![Site distribution map](figures/site_distribution_osm.png)
+
+### 2) Per-site class pixel-count heatmap
+![Per-site class heatmap](figures/site_class_pixel_counts_heatmap_merged.png)
+
+### 3) Train/val/test class distribution
+![Split class distribution](figures/train_val_test_class_distribution_merged.png)