A recent announcement on LinkedIn by Michael Egan, one of the co-founders of Bunting Labs, caught my attention the other day. Bunting Labs has released the new QGIS AI Map Tracing Plugin, an artificial intelligence-driven automatic digitization tool for features from scanned maps and plans. Egan and co-founder Brendan Ashworth developed this plugin to speed up the process of extracting geographic features from scanned maps and plans.
Extracting geospatial information from a PDF typically requires a tedious process of manually tracing features such as lines, polygons, and points from the PDF or scanned map image to create GIS vector data. . Depending on the complexity of your PDF or scanned map, this process can be very time-consuming.
After six months of development and 158,000 map training data points, Bunting Labs has released a map tracing plugin to automate this process. This plugin is designed to replace QGIS’s Add Line/Polygon Features or Create Features tools that GIS users typically use for heads-up digitizing. Instead, users seed the digitization process by tracing the beginning of a line or polygon and let the plugin’s autocomplete feature take over.
Given the great value that plugins like this offer for quickly extracting data from scanned maps and PDFs, I wanted to test it out on a sample map. Users can use her QGIS AI Map Tracing Plugin as a demo. If you provide your work email, you can test 2,000 completions, which Bunting Labs defines as “all line segments (additional vertices) added to a line or polygon after a click.” .
How to install the QGIS AI map tracing plugin
This plugin is available in the QGIS plugin repository ([プラグイン]->[プラグインの管理とインストール]) or by visiting the Bunting Labs site and uploading to QGIS.
Once you’ve installed the plugin, you’ll be able to redeem your 2,000 completion credits by registering your work email address. By registering your work email address, you will have access to a “Plugin Private Key” that allows you to use those credits.
Test the digitize plugin using some sample maps
To test the functionality of the plugin, you need to load digital files such as GeoTIFF, TIFF, PDF, or other image formats directly into QGIS. Serving images from a web service/REST server does not work. For this review, I georeferenced a historical map for testing purposes, but if you don’t want to georeference the scanned map yourself, you can use a map with the geographic coordinates already attached, like this collection of maps from the David Rumsey Map Collection. Search for historical maps.
The historical map I chose is an 1896 scanned map showing property owners in Clay County, Iowa. (As an aside, I chose this map in part because the cartographer who created it couldn’t even be bothered to spell the counties correctly).
Map typos aside, this map represented a good example of the condition of an old scanned map, containing the kinds of features a cartographer might want to extract from a historical map.
Automatically complete digitization of features from maps using AI Vectorizer
Using the AI digitization plugin is very easy. After adding the georeferenced map to the QGIS project, I set up a new data layer with polygons as a feature class so that I could test the functionality of the plugin when digitizing one of the lakes on the Clay County map.
I focused on digitizing Swan Lake, which is located in the top right quadrant of the map.
Following Bunting Lab’s instructions, I first turned on editing mode for the newly created polygon shapefile by clicking on the pencil icon. When editing mode is turned on, the AI Vectorization tool button changes from grayed out to active.
The AI Vectorization tool eliminates the need to trigger the Add Line/Polygon Features or Create Features tools that you would normally use to start digitizing features. Clicking “Vectorize with AI” will launch the tool and the cursor will change from a black arrow to a target. The tool itself doesn’t seem to have any settings. Simply trigger the tool and set the seed for digitization by tracing a small section of the feature you want to capture.
To begin the AI digitization, I clicked three times along the polygonal feature of the lake I wanted to capture and moved the cursor inside the polygon. It’s really satisfying to see the plugin in action. In most cases, the plugin was able to trace the boundaries of the lake cleanly and input far more vertices than manual digitization. If you want to see this automatic digitization in action, Bunting Lab has a short clip showing it in action.
There was one section where the AI plugin strayed along a property line adjacent to a lake.
According to Bunting Labs, the solution to incorrect digitization is to hold the mouse over the section where the digitization went off course, click the Shift key, and then left-click the mouse to cut that area. I found it very easy to fix over-digitization by following these instructions, even if you’re limited to using a trackpad.
If you hold down the Shift key, the last vertex will be highlighted in teal, making it easy to hover over it and click to delete it. After removing the incorrect digitization, I was able to manually click new vertices in the problem area to move the AI plug-in past the complexity and restart the vectorizer automation.
The end result of digitizing Swan Lake from the map was pretty decent. The vertices digitized by the AI tool were cleaner than those created by manual digitization. There were still some jagged sections where I stepped into manually moving the tool over some tricky parts.
It took some trial and error to get used to the functionality of this tool. I found that if I didn’t use the shift key quickly enough to pause the AI digitization, it would create a lot of erroneous vertices, and then trying to fix them would create weird polygons. I found it easier to delete the feature and try again, which causes the AI tool to pause before too many vertices are digitized.
Overall impression
This tool is very attractive considering its promise. Testing the capabilities of the AI vectorization tool definitely sped up on-screen digitization by quickly identifying and vectorizing features from scanned maps. This is completely inaccurate, as older maps can have a lot of inherent noise due to age and the scanning process, and of course still have a significant amount of misdigitization. It is not a manual process.
You may find more success if you actively guide the tool by manually adding vertices where the tool has previously been known to malfunction and start digitizing features outside of where you want them to digitize. got it.
It was relatively easy to pause the AI digitization using the Shift key and manually remove the erroneous vertices. Depending on the level of spatial accuracy, the final vector product may need to be manually cleaned to adjust for areas of features that deviate from the original source map.
In some situations, the tool would go far off track and jump to features outside the current view of the map canvas, even though none of the features were near the features you were trying to digitize. There are no settings to set tolerance levels or limit the tool to specific areas of the map. Adding these settings may reduce user frustration in this situation.
Overall, this tool is definitely something you’ll want to add to your toolkit for extracting features from scanned maps. Even with the incorrect digitization, with some manual adjustments from the yellowed historical map, we were able to vectorize the lake polygons faster and cleaner.
The 2,000 completion credits included with your free trial will expire quickly. I quickly ran out of credits while testing the plugin by trying out both the polygon and line features. If you only have one or two maps per month where you want to digitize some features, the free trial allows up to 1,500 completion credits per month.
For users with heavy digitalization workloads, two price ranges. At both price points, you get access to faster, more accurate AI models and priority support. The top tier (currently $99 per month) also offers user-provided map training to further improve the AI model. Bunting Labs also offers her 20-minute free demo session with the developer.
