Journey to YUNO 06|Did They Really Eat? YUNO Helps You Know
When we are away from home, the biggest feeding question is often not whether the feeder dispensed food. It is whether your pet actually came to eat, how much food is still in the bowl, and whether the next scheduled feeding is really needed.
Many automatic feeders can tell you that food has been dispensed. But that does not always mean your pet has actually eaten. It also does not tell you how much food is still sitting in the bowl.
With YUNO, we wanted the camera and AI to understand more of the real feeding process. The system is designed to recognize food dispensing, and also detect when a cat approaches the bowl and starts eating.
YUNO is designed to recognize both food dispensing and eating behavior, so owners can check more than a basic feeding log.The Problem With Leftover Food
There was another important challenge: how can YUNO understand how much food is still left in the bowl?
If a large amount of food is already remaining, continuing to dispense more can cause fresh kibble to pile on top of older kibble. Over time, food may stay exposed to air for longer, which can affect freshness, texture, and crunch.
That is why we developed YUNO's leftover food detection algorithm.
How YUNO Reads the Bowl
Powered by a MobileNet-based vision model, YUNO's leftover food detection algorithm is designed to recognize different levels of remaining food in the bowl: full, half, low, and empty.
YUNO reads the bowl as part of the real feeding process, not as a single yes-or-no food check.Training for Real Bowls, Real Food, Real Homes
Real homes are not controlled lab environments. Some users may use stainless-steel bowls. Others may use ceramic bowls or darker bowls. Kibble may be yellow, brown, dark, small, large, round, or irregular. The bowl may also contain crumbs, residue, or long-term use marks.
If an algorithm only works under ideal conditions, it cannot truly support everyday pet care. To improve robustness, we trained and tested YUNO's leftover food detection with a wide range of real-world data, including:
- Different bowl shapes, colors, and materials.
- Different kibble colors, sizes, and textures.
- Bright daylight, dim indoor light, and night-vision conditions.
- Clean bowls, bowls with crumbs, and bowls with long-term use residue.
Full-bowl examples help YUNO understand when enough food is already left.
Half-level examples help YUNO recognize moderate leftover food across different conditions.
Low-level examples help distinguish scattered kibble from an almost empty bowl.
Empty-bowl examples include clean bowls, crumbs, residue, and long-term use marks.From Dispensing Food to Understanding Feeding
In our internal test dataset, YUNO achieved over 99% accuracy in recognizing leftover food levels across tested bowl types, kibble types, and lighting conditions.
This allows YUNO to do more than simply check whether food exists. It helps the feeder better understand the bowl's real status and reduce unnecessary dispensing when there is already enough food left.
For users, this means fewer repeated feedings, less food sitting exposed for too long, and more confidence when checking in from afar.
For pets, it means a more thoughtful and consistent feeding experience.
YUNO Does Not Just Feed
YUNO was not designed only to answer, "Did the feeder dispense food?" It was designed to help answer, "Did my pet come to eat?", "How much food is still left?", and "Is the next feeding really needed?" Because truly smart feeding is not just about releasing food on schedule. It is about understanding what happens before and after every meal.
Leave a comment