State-of-the-art computer vision, without months of development.
Ocelo is the complete platform to experiment with, develop and put into production visual inspection models: from the first photos to a model monitored on the line, with the best open models and without writing code. Less time, less cost, and results you can prove.
Real Ocelo video: 1,760 photos of eight products; Ocelo groups the near-duplicates so you decide which ones stay.
Real Ocelo screenshots and videos, with public datasets of wood, textiles, concrete, food and electronics. Every step in detail
Proven models and open formats
DINOv2
/
ConvNeXt
/
EfficientNet
/
ResNet
/
MobileNetV3
/
RF-DETR
/
U-Net
/
FPN
/
SAM 2.1
/
ONNX
/
Docker
/
COCO
/
YOLO
/
Pascal VOC
/
PNG masks
/
DINOv2
/
ConvNeXt
/
EfficientNet
/
ResNet
/
MobileNetV3
/
RF-DETR
/
U-Net
/
FPN
/
SAM 2.1
/
ONNX
/
Docker
/
COCO
/
YOLO
/
Pascal VOC
/
PNG masks
/
Why Ocelo
Everything a vision project needs, in one platform
A computer vision project is usually spread across separate labeling tools, training scripts, servers to maintain and a team to hold it all together. Ocelo brings it into one place and applies what works best today at every step.
State-of-the-art results
Reference models such as DINOv2, ConvNeXt, RF-DETR and SAM 2.1, ready to train on your images. Ocelo proposes the model, the augmentation recipe and the input size for your case, and explains why.
Less development time
From labeling to deployment without switching tools or writing code. A first baseline in seconds, and every experiment saved with its data, its recipe and its environment, to compare and repeat it.
Lower cost
No infrastructure to build and no MLOps team: a fixed-price plan with cloud GPUs for training, backups and updates included. The model runs on a PC at the plant.
Results you can defend
The data checks review the data before training, the test set is locked, and every model comes with its acceptance report: you know how accurate it is and where it fails before it reaches the line.
Ocelo learns from what your team marks on the images. If an expert can tell the defect in a photo, it is a candidate for automation: on the line, in the lab or in the clinic.
Automotive · DetectionScratches and dents on machined parts
Food · SegmentationBruises and rot on fruit
Textiles · DetectionHoles and flaws in fabric
Packaging · DetectionBottles with missing or loose caps
Ceramics · SegmentationCracks in tiles
Plastics · DetectionBurrs on injection-molded parts
Electronics · DetectionMissing or misplaced components
Metalworking · DetectionPores in weld seams
Ophthalmology · SegmentationFundus imaging: locating the optic disc
Hematology · SegmentationBlood smears: locating and counting white blood cells
Pharma · DetectionBroken tablets in blister packs
Pharma · DetectionVials with badly fitted caps
Illustrative images. Swipe to see more.
Trust
Your data and your models are yours
Your own instance, isolated from the others and hosted in the EU.
Images, labels and models in open formats: COCO, YOLO, PNG masks and ONNX.
A chained, verifiable audit log, and a card for every model.
Encrypted daily backups and a server with no ports open to the Internet.
Ocelo Local
When the images cannot leave the building
The same platform on your company’s servers: the data stays on your network, also during training, and it works without an Internet connection.
Startup has a fixed price and covers a complete case, from the images to the model deployed on the line. For several lines, plants or an installation on your servers, let’s talk.