Image Segmentation
Image segmentation classifies every pixel in an image, assigning each to a class. It provides fine‑grained understanding: where exactly the object boundaries are.
Types of Segmentation
- Semantic segmentation: Assigns a class to each pixel (e.g., all cars are class 'car', all roads 'road').
- Instance segmentation: Differentiates individual objects (car1, car2, car3).
- Panoptic segmentation: Combines semantic and instance (everything is labeled, instances are separated).
U‑Net
U‑Net is a popular architecture for medical image segmentation. It has an encoder (downsampling) and decoder (upsampling) with skip connections to preserve fine details. Works well with limited training data.
Mask R‑CNN
Extends Faster R‑CNN by adding a mask prediction branch for each detected object. Outputs: bounding box, class label, and pixel‑level mask. State‑of‑the‑art for instance segmentation.
Fully Convolutional Networks (FCN)
The first end‑to‑end network for semantic segmentation. Replaces dense layers with 1×1 convolutions to output spatial maps. Upsamples using transposed convolutions.
Two Minute Drill
- Semantic: class per pixel; instance: separate objects.
- U‑Net: popular for medical segmentation.
- Mask R‑CNN: instance segmentation.
- FCN: first end‑to‑end semantic segmentation.
Need more clarification?
Drop us an email at career@quipoinfotech.com