Product
MASETOOL
AI model training for custom computer-vision projects. Describe what to detect and where; MASETOOL builds the dataset, trains a detector and packages it for your hardware.
Who it is for
Teams that need a detector for their own civil use case.
Inspection and monitoring
Infrastructure inspection, wildlife and fish surveys, environmental monitoring and airspace safety around airports.
Edge hardware
Models are sized for the device you name: NVIDIA Jetson Orin Nano 8 GB, NVIDIA Jetson Orin NX 16 GB, NVIDIA Jetson AGX Orin 64 GB, Raspberry Pi 5, PC, CPU only (x86), PC with NVIDIA GPU (x86).
Workflow
From a short form to a model package.
Describe the job
Pick the classes to detect, the camera, distances, light, weather, backgrounds and the computer the model will run on.
See if it is trainable
MASETOOL checks the request as you type and shows the training image size, model size and a planning estimate of the training time.
Add your own images (optional)
Upload labelled images (LabelMe / X-AnyLabeling JSON or YOLO). Every label is checked; problems are listed per file.
Train and download
Follow the job live. When it is done, download the model package with its report, results on the recorded test set and checksums.
Guided input
No requirement file to edit.
The wizard asks for each value with its unit and accepts how you naturally type it. MASETOOL’s own checks run before anything is trained, and its documented messages are shown next to the field they concern.
- Classes from the MASETOOL catalogue, or new classes reviewed by our team before training.
- Acceptance target as mAP50, mAP50-95, Recall, Precision.
- Colour, monochrome and thermal cameras.
Plans
Licensed separately from MADETOOL.
MASETOOL Professional is USD $40 per month. Enterprise is quoted per organisation. A MADETOOL plan does not include MASETOOL.
Compare MASETOOL plansScope
- Civil use only. People, faces, licence plates, weapons and military objects cannot be trained.
- Object detection only in this version; classification and segmentation are not trained yet.
- Results describe the recorded test set of each run, not performance in every environment. Training times are estimates.
- Latency is reported only when measured on the stated hardware.