YOLO26n room dirtiness
room dirt and clutter detector
1 min readupdated
This model is a fine-tuned Ultralytics checkpoint designed to detect five types of messes: liquid spills, trash cans, scattered litter, mold, and stains. Based on the lightweight YOLO architecture, it was trained on a specialized dataset of 1,200 images. The tool is well-suited for developers of smart home systems, cleaning robotics, or automated indoor cleanliness monitoring services.
Key features
- Fast inference with an average latency of 18.5 ms per image
- High overall accuracy with an mAP50 metric of 0.946
- Compact size: around 2.5 million parameters
- Support for detecting five distinct clutter categories
What you can do
- Integrate the model into robot vacuums or autonomous cleaning equipment to locate trash
- Set up video surveillance systems for automated room cleanliness monitoring
- Use in analytics applications to assess indoor dirt levels
Example
Example code for loading and running detection using the Ultralytics library:
from huggingface_hub import hf_hub_download
from ultralytics import YOLO
path = hf_hub_download(repo_id="tomas478/room-dirtiness-yolo26n", filename="best.pt")
model = YOLO(path)
results = model.predict("image.jpg", imgsz=640)
Read more
DetailsTest set metrics: mAP50-95 is 0.705, Precision is 0.913, Recall is 0.903.
Training: base model yolo26n.pt, 40 epochs, image size 640, AdamW optimizer.
Training hardware: NVIDIA RTX PRO 4000 Blackwell.








Comments