Overview

The Computer Vision Maps Module, part of the Data Augmentation suite, generates spatial context from raster data by quantifying local characteristics around each pixel.

It uses a pre-trained deep learning model, ResNet-50, to extract numerical features that describe textures, contrast, and variation in the surrounding area of the data. These features provide spatial context that goes beyond individual pixel values.

The module is commonly used with geophysical rasters, such as magnetics, and can also be applied to other raster datasets with uniform coverage and relevance to exploration.

Note: Access to Data Augmentation Modules is not available on all accounts. Contact your dedicated DORA contact to learn more.

What This Module Does

The Computer Vision Maps Module divides an input raster into overlapping patches and applies ResNet-50 feature extraction to each patch.

Rather than classifying the raster, the model returns numerical embeddings that summarize how the data varies locally. A single input raster can generate multiple output rasters, each representing a different extracted spatial feature.

These outputs are then used as input features in DORA’s Prediction Map workflow.

Learn more about Computer Vision Maps in this overview article.

Step-by-Step Instructions

Open Data Augmentation
- Once you've completed Step 3: Select Deposit Type, you can use the Data Augmentation Modules within an experiment.
- Click Data Augmentation in the top middle navigation bar:
Select Computer Vision Maps
- Click Computer Vision Maps, then Select.
Select a Grid Raster File
- Select the raw raster you want to process (for example, a magnetic grid).
- The module operates on raster data only and evaluates spatial patterns in the input grid, not individual data values in isolation.
Set the Resolution
- Set the output grid resolution.
- DORA displays the resolution of selected rasters to help you choose a matching size.
Clip Outliers (Optional)
- Toggle Clip Outlier on to remove extreme values and improve contrast.
- This can help when a small number of very high or low values dominate the colour range.
Apply Histogram Equalization (Optional)
- Toggle Histogram Equalization on to spread values across the colour range and improve contrast.
- This option enhances subtle variations in the raster and can make spatial patterns more visible.
Set the Number of Maps
- Set the number of output maps to generate.
- Each output map represents a different spatial feature extracted from the same input raster. A single input raster can produce multiple output layers.
- We recommend generating 3 or 4 as a starting point; higher numbers tend to repeat information.
Generate Computer Vision Maps
- Click Generate to run the feature extraction and create the output maps.
Review the Output Maps
- Review each output map and look for patterns that are consistent across maps.
- Watch for repetitive straight or grid-like lines, which can indicate artifacts from the original gridded dataset rather than meaningful spatial patterns.
Save or Export Result
- Rename the output raster if needed.
- Click the output to select it, then:
  - Export to download a .zip file of the raster.
  - Add to Input Features to use the map in the current experiment.

Interpreting the Output

Each output map represents a different spatial feature extracted from the same raster.

The outputs are not intended to be visualized or interpreted individually. Their value is in how they contribute spatial context when used as input features in DORA.

If repetitive, straight, or grid-like lines are present, these are often artifacts from the original gridded dataset rather than geological features. If artifacts persist, adjust the resolution or reprocess the source raster.