AutomaticDamageDetection
AI that finds damage in inspection photos
AI models that locate cracks, damp or spalling in images and draw their mask pixel by pixel.
Data processing and analysis
A web application that puts artificial intelligence models, trained for each type of damage, in the hands of any technician. It speeds up asset inspection and makes reporting easier: it returns a mask showing where the damage is, plus statistics for each image.
What it does
01Automatic masks
- Models trained to generate damage masks.
- Automatic identification of the selected damage type.
- Images of any size: they are tiled automatically to fit the model.
02Interactive validation
- Each mask is marked as correct or edited by drawing on the image.
03Results
- Final PNG mask with pixel-to-pixel correspondence with the original image.
- Statistics: area affected by each damage type and model used.
04Web interface
- Upload, processing and results in the browser.
- Status of each image at all times.
- Autosave so you can pick up where you left off.
Locate, identify and draw
The models perform three tasks: they locate where there is damage, identify its type (damp, crack…) and draw the exact mask of the pixels that make it up. Combining all the damage gives a complete mask overlaid on the image.
Fine-tuning
We start from pre-trained general models: accurate results with less training time and much smaller datasets.
Retraining
Masks validated by experts feed new training runs. The more it is used, the better it detects.
Cloud architecture
Cloud back-end, almost entirely serverless: full availability, the right hardware for the computation and scalability.
Where we have applied it
Want to see itin action?
We will show you Automatic Damage Detection with data and cases similar to yours, and answer your technical questions.