Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Webinar Agenda
Topic:
• Demystifying Deep Learning
• NVIDIA Tools & SDKs
• Deploying with Jetson
• Deep Vision Primitives
• 2 Days To A Demo
• Reinforcement Learning
• Simulation
• Conclusion / Q&A
“ Billions of intelligent devices will take advantage of DNNs to provide personalization and localization as GPUs become faster and faster over the next several years.”
— Tractica
INTELLIGENT MACHINES
Train Model
Robust Validation
Configure Network
Turing-Complete
Runtime Inference
Onboard Intelligence
A NEW COMPUTING MODEL
ERA of LEARNING
Deep Learning
Power Efficiency
Krizhevsky et al., NIPS 2012. ImageNet Classification with Deep Convolutional Neural Networks.
Image Classification Rate
Microsoft “Super Deep Network”
Berkeley’s Brett End-to-End
Reinforcement Learning
Deep Speech 2 One network, 2 languages
DeepMind AlphaGo Rivals a World Champion TU Delft Deep-Learning
Amazon Picking Champion
MILESTONES IN AI
Microsoft & Google “Superhuman” Image Recognition (ResNet)
WEpod Autonomous People Shuttle On the Road Starship Delivery
Robot Enters Service
2014-2016
PC GAMING
ONE ARCHITECTURE — END-TO-END AI
CUDA + Linux throughout the stack.
ONE ARCHITECTURE — END-TO-END AI
8
POWERING THE DEEP LEARNING ECOSYSTEM NVIDIA SDK accelerates every major framework
COMPUTER VISION
OBJECT DETECTION IMAGE CLASSIFICATION
SPEECH & AUDIO
VOICE RECOGNITION LANGUAGE TRANSLATION
NATURAL LANGUAGE PROCESSING
RECOMMENDATION ENGINES SENTIMENT ANALYSIS
DEEP LEARNING FRAMEWORKS
Mocha.jl
NVIDIA DEEP LEARNING SDK
developer.nvidia.com/deep-learning-software
DIGITS
Test Image
Monitor Progress Configure DNN Process Data Visualize Layers
Interactive Training with DIGITS
github.com/NVIDIA/DIGITS
github.com/NVIDIA/DIGITS
Classified Object!
Trained Deep Neural Net Model
Camera Inputs
X
Training: NVIDIA DIGITS Inference: NVIDIA Jetson TX1
TRAINING
DATA MANAGEMENT
MODEL VALIDATION
DGX-1
TESLA
TITAN-X
cuDNN TensorRT
DIGITS Workflow
JETSON TX1 EMBEDDED AI SUPERCOMPUTER
10W | 1 TF FP16 | >20 images/sec/W
JetPack 2.3
Software Components
Linux4Tegra R24.2 Ubuntu 16.04 LTS aarch64
Multimedia API SDK Low-level camera/codec
Runtime Inference TensorRT RC1 + FP16
CUDA Toolkit 8 cuDNN v5.1, cuGRAPH
VisionWorks 1.5.2.14 OpenCV4Tegra 2.4.13.17
OpenGL 4.5.0 EGL 1.4 / OpenGL ES 3.1
Tegra System Profiler Tegra Graphics Debugger
Double Performance, 20x Efficiency
developer.nvidia.com/jetpack
DIGITS OPTIMIZATION ENGINE
EXECUTION ENGINE
PLAN NEURAL NETWORK
TensorRT
• Fuse network layers
• Eliminate concatenation layers
• Kernel specialization
• Auto-tuning for target platform
• Select optimal tensor layout
• Batch size tuning TRAINED
NEURAL NETWORK
OPTIMIZED INFERENCE RUNTIME
developer.nvidia.com/TensorRT
Object Detection Localization
Image Recognition Classification
Segmentation Free Space
DEEP VISION
Registration SLAM
3D ShapeNet
GETTING STARTED JETSON COMMUNITY
Developer Forums devtalk.nvidia.com eLinux Wiki eLinux.org/Jetson_TX1
16
developer.nvidia.com/embedded
COMPREHENSIVE DEVELOPER S ITE
• JetPack SDK
• Libraries
• Developer tools
• Design collateral
• Developer Forum
• Training and Tutorials
• Ecosystem Partners
Hands-On Labs
nvidia.com/DLI
nvidia.com/DLI
Self-Paced Courses Nanodegree Programs
PARTNERSHIP Latest Technology GPU/DL Experts Global Sales & Marketing Network Training from Deep Learning Institute
INCEPTION PROGRAM
CONNECTING WITH AI STARTUPS
www.nvidia.com/inception
2 DAYS TO A DEMO
Ten Steps to Deep Learning
1. Install DIGITS on PC
2. Get JetPack 2.3 & TensorRT
3. Build the Vision Source
4. Dig Into the Example Code
5. Classify Images with ImageNet
6. Run the Camera Recognition Demo
7. Re-train the Network with DIGITS
8. Locate Object Coordinates - DetectNet
9. Run the Live Camera Detection Demo
10. Re-train DetectNet with DIGITS
github.com/dusty-nv/jetson-inference makezine.com/projects/jetson
Project includes: 1000 types of objects
pedestrian DetectNet
Robot OS (ROS) nodes
HORUS SOCIAL INNOVATION
AWARD WINNER
BLIND
ASSISTANCE
DEVICE
AERIALTRONICS
CNN
CNN
Sensory Inputs
CNN
Online L
earn
ing
MOTION CONTROL | BEHAVIOR IMITATION
PATH PLANNING | AUTONOMOUS NAVIGATION
Actions
feature maps
confidence values
Pixels to Actions
backpro
pagati
on
Experience Actuator
CNN RNN/LSTM
Loss/Reward
Solver
REINFORCEMENT LEARNING
OpenAI Gym
Gazebo
GTA-V
Unreal4Torch
OpenRAVE
TORCS
Physical Intuition
SIMULATION
LSTM ACCELERATION
cuDNN v5 - GPU-accelerated RNNs/LSTM
Forms internal working memory cells
Partially-observable inputs/environments
Launch a 2D grid of RNN cells
Supports:
Uni/Bidirectional RNNs
Non-uniform length minibatches
Dropout between layers
Jetson TX1 Developer Kit
Jetson EDU Program available
www.nvidia.com/embedded
THANK YOU! Q&A: WHAT CAN I HELP YOU BUILD?
2 Days To a Demo github.com/dusty-nv/jetson-inference
Reinforcement Learning github.com/dusty-nv/jetson-reinforcement
Access these Slides github.com/dusty-nv/jetson-presentations
Post Discussion devtalk.nvidia.com/default/topic/969035
eLinux TX1 Wiki eLinux.org/Jetson_TX1
Related Documents
