Dataset preparation
The toolkit extracts frames, organizes datasets, keeps numbered versions, and helps select representative training images.
Compan-IA R&D · Prior field example
A toolkit for training and reviewing vision models
A workspace for preparing data, annotating images, training models, and comparing results, shown with an anonymized agricultural example.
The starting point
Keep data preparation and model review in one place
Vision projects require many connected tasks: extracting frames, annotating images, managing dataset versions, training, and checking the output visually. Moving between separate tools makes that work slower and harder to repeat.
Work delivered
The main parts of the project
Assisted annotation
It proposes boxes and polygon outlines that a person can correct and approve before training.
Training and comparison
Model versions and their visual results can be compared before the output is connected to an operational process.
What we can verify
The same workflow, tested on an agricultural use case
The example shown compares box detection and segmentation on images from a Québec agricultural project. The model had to distinguish and locate visible parts of the plants in field conditions; the client, equipment, and operating design remain undisclosed.
View the service