Platform · ML Vision

Upload an image. Let it see.
Classify, detect, segment.

ML Vision is a computer-vision world inside AIRaML — three tasks in one app. Classify an image across 1000 categories (and compare four backbones), detect and box every object, or segment the scene pixel by pixel — all on real ONNX models from one FastAPI backend.

3 tasks in one app4 classifier backbonesONNX on FastAPIFree to use

You upload — cat.jpg

ResNet50 predicts

tabby cat91%
tiger cat6%
Egyptian cat2%

An illustrative classification. Your real image gets ranked labels, detection boxes or a pixel-level segmentation overlay — your choice of task.

One app, three vision tasks

Classify, detect, segment

The three workhorse computer-vision tasks, all on real ONNX models from one backend. Each is live, and each states the models and label set behind it.

● Live

Classification

Upload an image and get the top predicted labels with confidence, from any of four ImageNet-trained backbones you can compare side by side.

1000 ImageNet classesMobileNetV2 · ResNet50 · SqueezeNet · GoogLeNet
● Live

Detection

Find and box every object in the image, each with its label and confidence — people, vehicles, animals, everyday objects.

80 COCO classesTinyYOLOv3
● Live

Segmentation

Label the scene pixel by pixel — sky, road, building, person, furniture — as a coloured overlay on your image.

150 ADE20K classesSegFormer-B0

How it works

Pick → upload → read

01

Pick a task

Choose Classification (what is this?), Detection (what's where?) or Segmentation (which pixels are what?).

02

Upload an image

Drop in an image or use a sample. It's resized to each model's input size automatically. For classification, pick a backbone — or run several to compare.

03

Read the result

Ranked labels with confidence (classify), boxes over the image (detect), or a coloured pixel overlay with a class legend (segment).

Honest scope

Real models, with the caveats stated

Three real vision tasks on real ONNX models — compact models on fixed label sets, built to show the shape of practical vision, not to be a specialised system.

LiveWhat it does

  • Classification across 1000 ImageNet classes
  • Four backbones to compare — MobileNetV2, ResNet50, SqueezeNet, GoogLeNet
  • Object detection over 80 COCO classes (TinyYOLOv3)
  • Pixel-level segmentation over 150 ADE20K classes (SegFormer-B0)
  • Real ONNX models on a FastAPI backend
  • One microservice shared with ML Unified + EDA

CaveatsWhat to keep in mind

  • Fixed label sets — only names things in ImageNet / COCO / ADE20K
  • Compact models tuned for speed on CPU, not maximum accuracy
  • Expect misses on small, crowded or unusual scenes
  • General-purpose — not medical, security or inspection grade
  • Uploaded images are processed for the result, not kept

See your first image.

Pick a task, upload an image, and read the result — no setup, free.

Open the platform