Scribd
Upload
7 views14 pages

Object Segmentation Presentation

Object segmentation involves pixel-level labeling of objects in images, with types including semantic, instance, and panoptic segmentation. Regression predicts continuous values like bounding box coordinates, while segmentation offers pixel-wise classification for finer localization. Supervised learning requires labeled datasets for high accuracy, whereas unsupervised learning utilizes unlabeled data for pattern discovery.

Uploaded by

basit
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
7 views14 pages

Object Segmentation Presentation

Object segmentation involves pixel-level labeling of objects in images, with types including semantic, instance, and panoptic segmentation. Regression predicts continuous values like bounding box coordinates, while segmentation offers pixel-wise classification for finer localization. Supervised learning requires labeled datasets for high accuracy, whereas unsupervised learning utilizes unlabeled data for pattern discovery.

Uploaded by

basit
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 14

Object Segmentation

Regression vs Segmentation –
Supervised and Unsupervised
Learning
What is Object Segmentation?
• Object segmentation is pixel-level labeling of
objects in an image.
• Types: Semantic, Instance, Panoptic
Segmentation.
• More precise than object detection (bounding
boxes).
Regression vs Segmentation
• Regression: Predicts continuous values (e.g.,
bounding box coordinates).
• Segmentation: Pixel-wise classification
(discrete labels).
• Segmentation provides finer localization than
regression.
Comparison Summary
• Regression: Output - (x, y, width, height), Used
in detection.
• Segmentation: Output - Pixel masks, Used in
semantic/instance segmentation.
• Models: YOLO (Regression), U-Net, DeepLab
(Segmentation).
Supervised Learning in Segmentation
• Uses labeled datasets (input + ground truth
mask).
• High accuracy but needs extensive annotation.
• Examples: Mask R-CNN, DeepLab.
Unsupervised Learning in Segmentation
• Learns from unlabeled data (no ground truth).
• Uses clustering, autoencoders, GANs.
• Examples: K-means, PCA + clustering, anomaly
detection.
Types of Machine Learning
• Supervised Learning: Learning from labeled
data to predict outcomes.
• Unsupervised Learning: Discovering patterns
in unlabeled data.
• Reinforcement Learning: Learning through
interactions and rewards.
• Self-Supervised Learning: Generating labels
from data to learn representations.
Supervised Learning Sub-types
• 🔹 Classification - Support Vector Machine
(SVM)
• - Example: Email spam detection
• 🔹 Regression - Linear Regression
• - Example: Predicting house prices
Unsupervised Learning Sub-types
• 🔹 Clustering - K-Means
• - Example: Customer segmentation
• 🔹 Dimensionality Reduction - PCA
• - Example: Image compression, visualization
• 🔹 Anomaly Detection - Isolation Forest
• - Example: Fraud detection
Reinforcement Learning
• 🔹 Q-Learning
• - Example: Game playing (chess, tic-tac-toe)
• - Learns from rewards and penalties
Self-Supervised Learning
• 🔹 SimCLR (Contrastive Learning)
• - Example: Learning from unlabeled images
• - Learns by comparing similar/different
samples
Types of Supervised Learning
• Classification: Assign class labels (e.g., cat,
dog).
• Regression: Predict numeric values (e.g.,
bounding boxes).
• Structured Prediction: Outputs structured data
(e.g., segmentation masks).
Types of Unsupervised Learning
• Clustering: Group pixels based on similarity.
• Dimensionality Reduction: Simplify features
for segmentation.
• Anomaly Detection: Identify unusual regions
(e.g., tumors).
Final Summary
• Regression predicts object locations;
segmentation identifies exact object shape.
• Supervised learning provides higher accuracy
but needs labeled data.
• Unsupervised learning works without labels
but may be less precise.

You might also like