0% found this document useful (0 votes)
26 views44 pages

Vision Processing Unit

Uploaded by

f2022332007
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
0% found this document useful (0 votes)
26 views44 pages

Vision Processing Unit

Uploaded by

f2022332007
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/ 44

Click icon to add picture

VISION
PROCESSING
UNIT
GROUP MEMBERS:
Zoha Khalid, f2022332007
Tayyaba Sikandar, f2022332012
Rayyan Ahsan,f2022332051
WHAT IS COMPUTER
VISION
AND
HOW IT
WORKS?
Detailed Workflow of the VPU:
• Image Preprocessing: ALU array applies ISP algorithms to
raw image pixels.
• PE Array Processing: Handles convolutional layers, pooling,
and activation.
• Row Processor Tasks: Manages fully connected layers and
RNN computations.
• Output Handling: Final processed data is stored in external
memory.
• Data Exchange: Utilizes compression for efficient storage
and transfer.

1
Key Components of the VPU:
• Processing Elements (PEs): Basic computation units for
image/video processing.
• PE Blocks (PEBs): Clusters of PEs with buffers and ALU
arrays.
• ALU Array: Performs arithmetic operations and initial image
processing.
• Row Processor: Handles MAC operations and enhanced ALU
tasks.
• Global Buffer: Central storage for input and intermediate
data.

1
Memory Architecture in the VPU:
• Global Buffer: Main on-chip memory for data caching and
prefetching.
• Mmems in PEs: Local storage for input data and ISP tasks.
• Vertical and Horizontal Buffers: Temporarily hold data for
convolution computations.
• Row and Sharing Buffers: Store weights and input data for
FC layers and RNNs.
• Double-Buffering Technique: Allows parallel data loading
and processing for efficiency.

2
Types of Parallelism in VPUs:
Data Parallelism:
• Filtering: Convolutional filters applied concurrently across
image regions.
• Transformations: Simultaneous scaling, rotating, and
translating of image areas.
• Feature Extraction: Parallel analysis of image patches using
techniques like SIFT and HOG.

2
Types of Parallelism in VPUs:
Task Parallelism
• Image Acquisition: Different components handle acquisition and
preprocessing simultaneously.
• Feature Extraction and Classification: Concurrent feature
extraction and categorization processes.
• Simultaneous Operations: Separate cores handle real-time
capture, feature extraction, and classification.
Pipeline Parallelism:
• Concurrent Stage Processing: Different stages of the pipeline
process separate frames simultaneously.
• Overlapping Stages: Preprocessing, feature extraction, and
classification stages overlap to enhance throughput.
2
Types of Parallelism in VPUs:
Specific Parallel Processing Techniques:
• Single Instruction, Multiple Data (SIMD): Executes a
single instruction across multiple data points simultaneously.
• Single Instruction, Multiple Threads (SIMT):Multiple
threads execute the same instruction on different data
simultaneously.
• Vector Processing: Operates on vectors of data in parallel,
performing operations like addition and multiplication.
• Parallel Pipelining: Processes different stages of an image
processing pipeline concurrently.

2
Applications and Use Cases
• Autonomous Vehicles:
Real-time object detection, lane detection, and navigation.
• Drones and Robotics:
Navigation, obstacle avoidance, and detailed inspections.
• Smart Cameras and Surveillance Systems:
Facial recognition, behavior analysis, and urban surveillance.
• Augmented Reality (AR) and Virtual Reality (VR)Image tracking, gesture
recognition, and immersive experiences.
• Healthcare:
Accelerated medical imaging analysis and remote patient monitoring.
• Retail and Marketing:
Customer behavior analysis and automated checkout systems.
3
Click icon to add picture
Real-World Scenarios with Performance Benefits:
Real-World Scenarios:
• Tesla's Autopilot:
VPUs enable real-time decision-making and enhanced
navigation in autonomous driving.
• Amazon Go Stores:
Checkout-free shopping experience powered by VPU-based
item tracking.
• Medical Diagnostics:
Faster and more accurate medical image analysis for early
disease detection.

3
Industries Commonly Using VPUs:
Industries Commonly Using VPUs:
• Automotive Industry:
Autonomous driving and advanced driver-assistance systems (ADAS).
• Consumer Electronics:
Smart cameras, AR/VR devices, and mobile phones.
• Healthcare:
Medical imaging and remote monitoring.
• Retail and Marketing:
Customer analytics and personalized marketing.
• Security and Surveillance:
Facial recognition and behavior analysis.
3
Performance Metrics:

3
Trade-offs Between Performance, Power
Consumption, and Cost:

3
FUTURE TRENDS AND DEVELOPMENTS:
Current Trends:

4
EMERGING TECHNOLOGIES AND ARCHITECTURAL INNOVATIONS:

4
Impact of Hardware Advancements on
Architecture

4
Impact of Software Advancements on
Architecture

You might also like