Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning...
Transcript of Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning...
![Page 1: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/1.jpg)
Automated Optical Inspection and Defect
Detection with Deep Learning
Harshita Bhurat
Product Manager – Image Processing and Computer Vision
![Page 2: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/2.jpg)
What is Automated Optical Inspection?
“ Automated optical inspection is the image-based or visual
inspection of manufacturing parts where a camera scans
the device under test for both failures and quality defects"
3
Machine Vision Visual Inspection
Automated Inspection
Automated Defect Detection
![Page 3: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/3.jpg)
Automatic Defect Detection
Defect Detection in Railway Components
Visual Inspection of Automotive Parts
Pipe
Assess Pipe Weld Damage at Power Plants
Customer References
4
![Page 4: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/4.jpg)
5
![Page 5: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/5.jpg)
Can you find the defective hex nut?
6
![Page 6: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/6.jpg)
7
![Page 7: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/7.jpg)
8
![Page 8: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/8.jpg)
9
![Page 9: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/9.jpg)
Finding Defective Hex Nuts
10
![Page 10: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/10.jpg)
Detecting Parts
11
![Page 11: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/11.jpg)
12
Defect Detection Workflow
Ground truth labeling
DATA PREPARATION
Data access and preprocessing
Simulation-based data generation
I Iteration and Refinement
AI MODELING
Model exchange across frameworks
Model design andtuning
Hardware-accelerated training
DEPLOYMENT
Edge, cloud, desktop
Enterprise Systems
Embedded Devices
Ground truth labeling
12
![Page 12: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/12.jpg)
13
Defect Detection Workflow
Ground truth labeling
DATA PREPARATION
Data access and preprocessing
Simulation-based data generation
I Iteration and Refinement
AI MODELING
Model exchange across frameworks
Model design andtuning
Hardware-accelerated training
DEPLOYMENT
Edge, cloud, desktop
Enterprise Systems
Embedded Devices
Ground truth labeling
13
![Page 13: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/13.jpg)
Data Access and Preprocessing – Common Challenges
14
How do I access large data that might not fit in memory?
How do I preprocess data and get the right features?
How do I label my data faster?
What if I have an imbalanced dataset or don’t have enough data?
![Page 14: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/14.jpg)
Data Access and Preprocessing – Common Challenges
How do I access large data that might not fit in memory?
15
![Page 15: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/15.jpg)
How do I load and access large amounts of data?
Loads image/signal data
into memory as and
when needed
Work with out-of-
memory numeric data
– Train deep neural
networks for
numeric arrays
…
>> imageDatastore
>> audioDatastore
>> fileDatastore
Custom Datastores also
available
Datastores Tall Arrays
Work with very large, tiled and
multi-resolution images
BigImage
Each red box is a 1024-by-1024 tile in the file.L1's dimensions = 29,600 x 46,000L2's dimensions = 14,800 x 23,000L3's dimensions = 7,500 x 12,000Rows = 29600Columns = 46000TileSizeIntrinsic = [1024 1024]ResolutionLevelSizes = [29600 46000
14800 230007500
12000]CoarsestLevel = 3FinestLevel = 1PixelSpacings = [1 1; 2 2; 3.947 3.833]
16
![Page 16: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/16.jpg)
Data Access and Preprocessing – Common Challenges
17
How do I preprocess data and get the right features?
![Page 17: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/17.jpg)
Data Access and Preprocessing – Common Challenges
18
How do I preprocess data and get the right features?
![Page 18: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/18.jpg)
Pre-processing Data – Registration Estimator App
19
![Page 19: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/19.jpg)
Pre-processing Data – Image Segmenter App
20
![Page 20: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/20.jpg)
Preprocessing Data - Apps
21
Color Thresholder Image Region Analyzer
![Page 21: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/21.jpg)
Pre-processing Data – Built-in Algorithms
imgaborfilt fibermetricimadjust
22
![Page 22: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/22.jpg)
Without pre-processing With pre-processing
Defect detection using AlexNet: Results with preprocessing
Peak of normal score shifts right(Sufficient margin from
abnormal units)
Without pre-processing
score
nu
m o
f el
emen
ts
Histogram of predicted scores
With pre-processing
score
Separated defective units clearly
23
![Page 23: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/23.jpg)
Data Access and Preprocessing – Common Challenges
24
How do I label my data faster?
![Page 24: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/24.jpg)
Data Preprocessing - Labeling
25
![Page 25: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/25.jpg)
Image & Video Labeler
Image Labeler
+ Video labeler
Big-Image
Labeler
26
![Page 26: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/26.jpg)
Big Image Labeler
Image Labeler
+ Video labeler
Big-Image
Labeler
27
![Page 27: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/27.jpg)
Data Access and Preprocessing – Common Challenges
What if I have an imbalanced dataset or don’t have enough data?
28
![Page 28: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/28.jpg)
Contrast Jitter on Grayscale
Data Augmentation by Transformation
Original
Dataset
Augmented
Dataset
N times as
much data
29
![Page 29: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/29.jpg)
Data Augmentation : Generative Adversarial Networks (GANs)
Generative Adversarial Networks
30
![Page 30: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/30.jpg)
31
Defect Detection Workflow
Ground truth labeling
DATA PREPARATION
Data access and preprocessing
Simulation-based data generation
I Iteration and Refinement
AI MODELING
Model exchange across frameworks
Model design andtuning
Hardware-accelerated training
DEPLOYMENT
Edge, cloud, desktop
Enterprise Systems
Embedded Devices
Ground truth labeling
31
![Page 31: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/31.jpg)
Deep Learning for Defect Detection
32
Deep learning for Classification
Deep Learning for Object Detection
![Page 32: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/32.jpg)
Deep Learning for Defect Detection – Multiple techniques
33
Deep learning for Classification
![Page 33: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/33.jpg)
Two Approaches for Deep Learning
2. Fine-tune a pre-trained model (transfer learning)
1. Train a deep neural network from scratch
34
![Page 34: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/34.jpg)
Train a Deep Neural Network from Scratch
35
![Page 35: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/35.jpg)
Convolution +
ReLu PoolingInput
Convolution +
ReLu Pooling
…
Feature Learning
…
Fully
ConnectedSoftmax
cartruck
bicycle
…
van
… …
Classification
Two approaches for Deep learning
Fully
Connected Softmaxfood
tree
…
New Classification
Approach 2. Fine-tune a pre-trained model (Transfer learning)
36
![Page 36: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/36.jpg)
Fine-tune a Pre-trained Model (Transfer Learning)
37
![Page 37: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/37.jpg)
Classification with Trained MobileNetV2
38
![Page 38: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/38.jpg)
Challenges with Deep Learning Models
Car
Truck
Bicycle
• Class Activation Mapping(CAM)• Grad-CAM
Explainable AI
is required
39
![Page 39: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/39.jpg)
Classified as “keyboard” due to
the presence of the mouse
Incorrectly classified “coffee
mug” as “buckle” due to the
watch
Class Activation Mapping to Investigate Network Predictions
40
![Page 40: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/40.jpg)
Visualization of Features with CAM
OK → Reacts to whole surface
Bad→ Reacts to the scratch
Captured Image Classification and CAM
41
![Page 41: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/41.jpg)
Deep Learning for Defect Detection
42
Deep Learning for Object Detection
![Page 42: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/42.jpg)
Detecting Objects with You Only Look Once (YOLO) v2
Predictions
YOLO
CNN
NetworkDecode
Predictions
Build, test, and deploy a deep learning solution that can detect objects in images and video
43
![Page 43: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/43.jpg)
Mask Detection with YOLO v2
44
![Page 44: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/44.jpg)
Experiment Manager
45
![Page 45: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/45.jpg)
Defect Detection Workflow
Ground truth labeling
DATA PREPARATION
Data access and preprocessing
Simulation-based data generation
I Iteration and Refinement
AI MODELING
Model exchange across frameworks
Model design andtuning
Hardware-accelerated training
DEPLOYMENT
Edge, cloud, desktop
Enterprise Systems
Embedded Devices
Ground truth labeling
46
![Page 46: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/46.jpg)
Deploy to Any Processor with Best-in-class Performance
FPGA
CPU
GPUCode
Generation
47
![Page 47: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/47.jpg)
Deploy defect detection algorithms from
MATLAB to Jetson AGX XavierDeploy defect detection algorithms from
MATLAB to ZCU102 board from Xilinx
Deploy to Hardware
48
![Page 48: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/48.jpg)
Deploy to Hardware
Resources:
▪ Deploying Deep Neural Networks to GPUs and
CPUs Using MATLAB Coder and GPU Coder
▪ Using GPU Coder to Prototype and Deploy on
NVIDIA Drive, Jetson
▪ Real-Time Object Detection with YOLO v2 Using
GPU Coder
▪ Image Classification on ARM CPU: SqueezeNet
on Raspberry Pi
▪ Deep Learning on an Intel Processor with MKL-
DNNDefect detection deployed on
ARM Cortex-A microprocessor
49
![Page 49: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/49.jpg)
Deploy to Enterprise IT Infrastructure
Request
Broker
MATLAB
Production Server
Custom Tools
Streaming
Dashboards
Databases
Cloud Storage
Containers
Cloud & Datacenter
Infrastructure
50
![Page 50: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/50.jpg)
Defect Detection Workflow
Ground truth labeling
DATA PREPARATION
Data access and preprocessing
Simulation-based data generation
I Iteration and Refinement
AI MODELING
Model exchange across frameworks
Model design andtuning
Hardware-accelerated training
DEPLOYMENT
Edge, cloud, desktop
Enterprise Systems
Embedded Devices
Ground truth labeling
51
![Page 51: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/51.jpg)
Key Takeaways
▪ Interactive and easy to use apps help explore, iterate and automate
workflows
▪ Flexibility and options to choose networks and optimizations based on
data and requirements
▪ MATLAB provides an easy and extensible framework for defect
detection from data access to deployment
![Page 52: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/52.jpg)
![Page 53: Automated Optical Inspection and Defect Detection€¦ · across frameworks Model design and tuning Hardware-accelerated training DEPLOYMENT Edge, cloud, ... Data Access and Preprocessing](https://reader033.fdocuments.in/reader033/viewer/2022043022/5f3d8a431a32fa446a61d5bc/html5/thumbnails/53.jpg)
THANK YOU!
54