PerfectWin commited on
Commit
310857f
·
verified ·
1 Parent(s): bb66a7f

Card v0.7: realistic naming results, new files

Browse files
Files changed (1) hide show
  1. README.md +15 -6
README.md CHANGED
@@ -15,7 +15,7 @@ tags:
15
 
16
  # GeoRas2Vec model – scanned geological map → GIS polygons, faults and unit names
17
 
18
- **Research preview (v0.6).** **Code, installation and documentation: [GeoRas2Vec on GitHub](https://github.com/PerfectWin7777/georas2vec).**
19
 
20
  GeoRas2Vec turns a scanned or rasterised geological map into a GeoPackage that opens directly in QGIS or ArcGIS. The GeoPackage holds three things:
21
 
@@ -36,6 +36,9 @@ This repository holds the model files. The command-line tool downloads them auto
36
  | `naming_model.pkl` | scores (text read on the map, polygon) pairs |
37
  | `legend_ranker.pkl` | ranks the candidate codes of each polygon (text evidence, leader lines, legend swatch) |
38
  | `prop_model.pkl` | decides when a code read on the map can be propagated to polygons of the same colour |
 
 
 
39
  | `water_model.pkl` | recognises water / outside-the-map regions |
40
  | `exemple/` | example output GeoPackage, legend CSV, QGIS render |
41
  | `training/` | training curves and metrics of the network |
@@ -71,10 +74,15 @@ The ranker and the propagation model were trained without this region.
71
  | names from text only | 65.9 % | 94.1 % | 62.0 % |
72
  | **+ colour propagation** | **74.7 %** | **93.7 %** | **69.9 %** |
73
 
74
- > **Idealised legend.** These naming results use legend colours measured on the scan itself (close to a perfect
75
- > legend). With the printed legend read automatically, about 19 % of the area is correctly named from text and
76
- > legend (91–94 % precision, cross-sheet, before propagation). Matching printed swatches by colour alone is the
77
- > current bottleneck.
 
 
 
 
 
78
 
79
  ![Names](https://huggingface.co/PerfectWin/georas2vec-model/resolve/main/figures/naming_region.png)
80
 
@@ -93,7 +101,8 @@ The ranker and the propagation model were trained without this region.
93
  ## Versions
94
  | version | change |
95
  |---|---|
96
- | **v0.6** | English attribute table by default (`--lang fr` for French), age shown once, fragments < 200 px merged into their neighbours (3× fewer polygons), no colour propagation onto white paper / ice |
 
97
  | v0.5 | two-scale mutex watershed; fault lines split at junctions; topological contours (no gap, no overlap); water detection; OCR restricted to the legend alphabet; decoding of age prefixes, Ʀ, Є and superscripts; colour propagation over the whole map; block processing of large maps |
98
  | v0.2 | first public release (`revision="v0.2"`) |
99
 
 
15
 
16
  # GeoRas2Vec model – scanned geological map → GIS polygons, faults and unit names
17
 
18
+ **Research preview (v0.7).** **Code, installation and documentation: [GeoRas2Vec on GitHub](https://github.com/PerfectWin7777/georas2vec).**
19
 
20
  GeoRas2Vec turns a scanned or rasterised geological map into a GeoPackage that opens directly in QGIS or ArcGIS. The GeoPackage holds three things:
21
 
 
36
  | `naming_model.pkl` | scores (text read on the map, polygon) pairs |
37
  | `legend_ranker.pkl` | ranks the candidate codes of each polygon (text evidence, leader lines, legend swatch) |
38
  | `prop_model.pkl` | decides when a code read on the map can be propagated to polygons of the same colour |
39
+ | `code_reader.pt` | reader of geological unit codes (Є, Ʀ, superscripts, small capitals, offshore `*`) |
40
+ | `swatch_enc.pt` | learned legend-swatch / polygon comparator (colour + pattern) |
41
+ | `legend_ranker_sw.pkl` | naming ranker trained with real printed legends (used by v0.7) |
42
  | `water_model.pkl` | recognises water / outside-the-map regions |
43
  | `exemple/` | example output GeoPackage, legend CSV, QGIS render |
44
  | `training/` | training curves and metrics of the network |
 
74
  | names from text only | 65.9 % | 94.1 % | 62.0 % |
75
  | **+ colour propagation** | **74.7 %** | **93.7 %** | **69.9 %** |
76
 
77
+ > **Idealised legend.** The table above uses legend colours measured on the scan itself (close to a perfect legend).
78
+
79
+ **Realistic setting (v0.7)**: printed legend read automatically (swatches + codes), polygons predicted, 4 held-out zones
80
+ of the southern sheet, models trained on the northern sheet only.
81
+
82
+ | | text + legend | + colour propagation | names correct |
83
+ |---|---|---|---|
84
+ | real legend, colour signatures only | 17.2 % | 23.6 % | 90.1 % |
85
+ | **real legend + learned swatch comparator (v0.7)** | **31.8 %** | **37.5 %** | **87.6 %** |
86
 
87
  ![Names](https://huggingface.co/PerfectWin/georas2vec-model/resolve/main/figures/naming_region.png)
88
 
 
101
  ## Versions
102
  | version | change |
103
  |---|---|
104
+ | **v0.7** | automatic legend reading (swatch boxes + codes), geological code reader (56/60 real legend boxes vs 26/60 with generic OCR), learned swatch comparator, ranker trained with real printed legends |
105
+ | v0.6 | English attribute table by default (`--lang fr` for French), age shown once, fragments < 200 px merged into their neighbours (3× fewer polygons), no colour propagation onto white paper / ice |
106
  | v0.5 | two-scale mutex watershed; fault lines split at junctions; topological contours (no gap, no overlap); water detection; OCR restricted to the legend alphabet; decoding of age prefixes, Ʀ, Є and superscripts; colour propagation over the whole map; block processing of large maps |
107
  | v0.2 | first public release (`revision="v0.2"`) |
108