Research Paper Deep Dive - Vision GNN: An Image is Worth Graph of Nodes

The deep learning research team from Huawei proposed a new Deep Learning Graph Neural Network Model (ViG GNN) to perform deep learning image recognition and object detection task using GNN or Graph Neural Networks.

The proposal is to split the source images in to equal size patches first and build a graph neural network from each image patch and then apply a combination of Graph Processing and Feed Forward Network to build the Graph Neural Network which can surpass existing CNN and GNN models for image recognition and object detection.

In this video we are taking a deep dive to learn the more about the ViG model implementation details and how does it is different from the Google ViT model.

GitHub Resources:

Research Paper and Code:

▬▬▬▬▬▬ ⏰ TUTORIAL TIME STAMPS ⏰ ▬▬▬▬▬▬
- (00:00) Paper Introduction
- (01:19) Suggested GNN Tutorials
- (02:05) Deep Dive Starts
- (03:08) Graph Representation of Images
- (05:35) Graph Processing
- (07:26) ViG Block
- (07:58) Construction Graph Structure
- (09:11) ViG Isotropic and Pyramid CV Architecture
- (10:32) Research Paper and GitHub Code Reference

Connect
------------------

Tags:
#gnn #ai #cnn #ml #lime #aicloud #h2oai #driverlessai #machinelearning #cloud #mlops #model #collaboration #deeplearning #modelserving #modeldeployment #pytorch #datarobot #datahub #streamlit #modeltesting #codeartifact #dataartifact #modelartifact #onnx #aws #kaggle #mapbox #lightgbm #xgboost #classification #dataengineering #pandas #keras #tensorflow #tensorboard #cnn #prodramp #avkashchauhan #LIME #modelexplanations #mli #xai

[mertardaasar]

Your explanation is very clear, please post more paper explanation videos about graph neural networks.

[jihunbae]

I finally found a good paper and implement code. great job and really nice presentation. Thx a lot !

[feltonjackson]

Great work. It was helpful to learn the technology without the complexity of math behind it, keep it coming.

[pankajchand]

How is each patch made into a feature vector? What is the feature extraction process?

[Alohahola]

Thanks sir for your explanation, i'm just asking if we really need this, (representing images as graphs) and do we need tools or libraries that are cencerned with this transfornations (image 2 graph) for further applications (GNN, ...), Thanks again for your help.

[kounisameh]

great work, Can you explain how to generate graph data for a custom image database in simple coding?

[EWelt-sxkj]

Thank you for the nice explanation. It is really helpful.
And I have quick question: for each block, is a new graph with a different set of edges reconstructed according to the updated features?

[MdRakibulIslam-btui]

Hello sir, I watched your video related with GNN. Your videos help me a lot to understand GNN. My question to you is, is it possible to apply GNN for Medical Image Processing/Classification task?

[priyankakarthi]

Sir... How to implement this paper code? Please help me sir..

[dmitrymitrofanov]

thanks, great job.New chanel for my top list.