In the world that quickly gets up the deep learning and a computer view, some buildings have left a lasting impact due to their simple, efficiency and disability. One of such gregg model, developed by the Visual Geometry Group at University of Oxford.
If you examine Neal Rural Neal Neural Neural (CNN) networks or you are looking for a powerful, well-specified model of images, understanding the VGG must.
In this article, we will cover What is VGGits architecture, Benefits, GrimThe real world dogsand frequently asked questions to show a complete picture of why VGG continues to influence a deep reading today.
What is Visual Geometry Group (VGG)?
VGG, short The geometry groupIt is the last deeper stadiums used by neural and Convelleval neural arkys known their own layers. Name “Deeper” Say a large number of layers on a network, with Vgg-16 including Vgg-19 united 16 and 19 layloval layersrespectively.
VGG played an important role in improving Types of accreditation of object and issued many basic models for various activities and details, including Nqelenqup.
Though established years ago, there is always one of the most commonly used buildings of Recognition of images because of its operation and formal formation.
Why is VGG important?
VGG's success lies easily and applies:
- It uses 3 × 3 3 layers in addition to each other.
- Increases the depth to improve accuracy.
- It is highly referred to different tasks of vision such as the acquisition of an object, separation, and style transmission.
Or new buildings such as Break including Successful They passed a well-efficient VGG, VGG lives IA Basic model On the computer Vision Education and Practice.
Suggested: Free Lessons for Deep Learning Lessons
The construction of vgg is described in detail
This page The construction of vgg prominent because of its good convenience and formal formation. The main idea focuses on working Small Consculval filter (3 × 3) and set yourself deeply to capture complex features from idols.
Let's clear the status of the building by step:
1. Input layout:
- Input size: VGG is designed to take pictures of random size of 224 X 224 pixels have 3 channels (RGB).
Example: Input formation = (224, 224, 3)
2. Wishstallers:
- VGG uses The layers of a lot of size by:
- Sorting size: 3 × 3
- Stride: 1
- Padding: 'Same' (to keep local repairs)
- 3 × 3 kernel captures good details while putting the layers rise The field receives.
- Depth is increased Continuously to the network by adding multiple filters (from 64 and up to 512).
Why Stop many measurements 3 × 3 3?
To add two 3 issues
3. Working activity:
- After all the Conmotolal layer, VGG works a Push (A fixed unit) activated.
- This is introducing inequality, to help the network read clearly complex patterns.
Formula: Relu (x) = max (0, x)
4. Integration layers:
- Behind all a few special blocks, VGG uses a The Max Pooling layer.
- Sorting size: 2 × 2
- Stride: 2
- PURPOSE: To reduce the area of the area (height and width) while storing the most important factors.
- This helps reduce the integration and overrule.
5. Completely connected layers (combined):
- After working solution and Pooling, outgoing feels in 1D vector.
- VGG usually uses Two or three connected layers:
- The first layers of FC: 4096 Neurons
- The last FC layer: The number of neurons equal to the number of output phase (eg 1000 for ImageNet).
These layers serve as classifier over issued areas.
6.
- The last layer is using Softmax Activation work access to opportunities to get out of each class.
- For example:
- With Imaganet, predicts more Categories of an item 1,000.
7. parameter to count:
One of the vgg's biggest symptoms is a large number of parametersmainly as a result of a completely connected layer.
| Statue | Complete parameters | Breast |
| Vgg16 | ~ 138 Million | The Heart Rest |
| Vgg19 | ~ 143 Million | 19th |
This makes VGG It is very expensive But also too strong by reading rich feature representations.
VGG16 Architecture crack (for example):
| The type of layer | The size of the output | Filters / neurons |
| Install | (224, 224, 3) | – |
| It has always been up – 64 x2 | (224, 224, 64) | 64 |
| Max Pool | (112, 112, 64) | – |
| Convel3-128 X2 | (112, 112, 128) | 128 |
| Max Pool | (56, 56, 128) | – |
| ICONT3-256 X3 | (56, 56, 256) | 25 |
| Max Pool | (28, 28, 256) | – |
| Conf3-512 x3 | (28, 28, 512) | 512 |
| Max Pool | (14, 14, 512) | – |
| Conf3-512 x3 | (14, 14, 512) | 512 |
| Max Pool | (7, 7, 512) | – |
| Flatten | (25088 | – |
| Completely connected | (4096) | 4096 |
| Completely connected | (4096) | 4096 |
| Completely connected | (1000) | 1000 |
| Softmax releases | (1000) | – |
Why is VGG construction special?
- Divides us: It also is the same block structure, which makes it easy to measure and change.
- The lift feature: Lowly layers are reading simple aspects (edges, colors), while deep layers read the coins of the unity (shapes, objects).
- Transfers: The features learned by VGG is effective in different datasets, which is why VGG models are previously used in learning.
VGG Benefits
- Simple: The construction of the same vgg (the 3 × 3 filters) makes it easy to understand and use.
- Transfer a friendly learning: VGG models are previously used for extension, time savings and resources in new projects.
- A Powerful Foundation: Despite being older, VGG works as a solid basis for a number of research and applications.
- A consistent operation: VGG is reliably active in various visual activities to look at the picture.
The bad of vgg
- Model size: VGG requires important storage (more than 500MB), making it easier to be moved on mobile devices or edge.
- More heavy: The model has high memory use and slow times due to its depth and the number of parameters.
- Laugh of modern buildings: Models are like reset and mobilet to achieve the same accuracy or better with fewer and immediate parameters.
Get used to love Algorithms study.
The actual use of vgg
| Ranch | An example of the application |
| Health care | Photographic image analysis and diagnosis |
| Cars | Recognition of item in private cars |
| Security | Face Access and Programs to Be Accountable |
| Sale | Search for productivity and recommendation |
| Art & Design | Type Transmission and Image Development |
Top VGG projects based on VGG
- Photo Style Transfer – Using VGG layers to combine one picture style in the content of another.
- Lining feature – Verligh VGG as a more feature with complex pipes.
- The discovery of something – combined with regional networks in the region in activities such as Immediately r-CNN.
Find that a Neural Neural Network (RNN) It also works for the use of language similarities and forecasts.
Store
Building VGG is a basic part of a deep reading history. VGG, in its refined convenience and demonstrates efficiency, valuable information for anyone who checks the computer view.
Whether you improve the research project, you use reading, or trying to transfer style, the VGG provides a solid basis.
Build a Machine Local Work
Our Scientific Science and Machine and Machine program teaches you that Neveral Neakers network techniques with a vision of the computer.
Read from MIT's intelligence through the manuals and maneuvers of personalization to create practical solutions, which are conducted. Sign up now to awaken your work in AI and Real-World Innovation.
Frequently Asked Questions
1. Why are 3 × 3 filters used in VGG?
VGG uses 3 filters
2. How does VGG compare recharge?
VGG is simple but heavier. Resetch uses residual connections to train deep networks through better performance and efficiency.
3. Is VGG can be used for non-photo data?
VGG is prepared for photos, but its Convelval principles are sometimes redesigned to follow consecutive data such as sound or video.
4. VGG16 and VGG19 differ?
The main difference is lying in the depth of VGG19 you have three colors layers than VGG16, which promote slightly accuracy but increase integration.
5. Is VGG still in use today?
Yes, especially in education, research bases, and transmit literacy, even though modern buildings may be organized in production locations.
