Computer vision
Peso na Granja
Estimate weight from a photograph.
6.5 kg mean absolute error on validation
Computer vision contribution and paper co-authorship
Metric obtained on validation with the public PigRGB-Weight dataset. It is not a measurement across all field conditions.
THE DECISION THAT MATTERS
Distinguish evaluation from field use
Validation evaluates the model on the studied dataset. Its interpretation must preserve the data context and limitations, even when the application is already integrated into the product.
The challenge
Investigate how RGB images can estimate pig weight and support operational monitoring.
The work
I contributed to weight estimation using a neural network and transfer learning, integrated into Peso na Granja, an app for pen management, weighing and dashboards.
The result
The model achieved a mean absolute error of 6.5 kg on validation with the public PigRGB-Weight dataset. The research was co-authored and published in RECIMA21, volume 7, issue 8, in August 2026.
The metric refers to the validation set, not performance measured across all field conditions. The application's code is proprietary.
Explore the technical details
Weight regression from RGB images with ResNet18 and transfer learning. Paper: “Pig weight estimation from RGB images using convolutional neural networks and the PigRGB-Weight dataset” (original in Portuguese). DOI: 10.47820/recima21.v7i8.8496.