About

Copyright © 2025 Universiti Utara Malaysia.

Dual Rabbit Analyser identifies a rabbit's breed and estimates its weight from a live camera feed. It combines two fine-tuned models that run entirely in the browser using TensorFlow.js: a breed classifier and a weight regressor. No frame is uploaded, sent to a server, or stored — everything is computed on the device.

The development of this application is funded by the UUM Case Study Grant under grant numbers 21621 and 21622. It is a collaborative effort between researchers from UUM, UITM Arau, and Suffhil Garden Rabbit Farm (SGRF).

Contact: The Digital Agri : Smart Rabbit Group

Breed identification

The breed classifier distinguishes the following classes:

• Chinchilla Mix Breed
• New Zealand Black
• New Zealand White
• Unknown

Weight estimation

The weight result is an interval, not a single figure. It is shown as Weight est: 2.03 - 2.47 kg style, being the estimated weight plus and minus 0.22 kg. That figure came from a validation mean absolute error of 0.090 kg on the model's training dataset.

Treat it as an estimate, not a scale reading. The weight estimate is only shown once the breed has been identified with reasonable confidence, because the two models are meant to be read together.

Taking a good reading

Fill the frame with the rabbit's body, at roughly one arm's length, in reasonably even light. Hold the camera still while the estimate settles — the displayed figure is the median of the last three frames, so it takes a few seconds to stop moving.