Uploaded byadityasiwach20
22 views

Convolutional Neural Networks(CNN) and Computer Vision

This presentation provides an in-depth overview of Computer Vision — a field of Artificial Intelligence that enables machines to interpret and understand visual information from the world. It begins by explaining the fundamental concepts of image processing, including image representation, filtering, and feature extraction. The presentation then transitions into Convolutional Neural Networks (CNNs), the backbone of modern computer vision systems. It explains the architecture of CNNs, covering layers such as convolution, pooling, activation, and fully connected layers, and how they work together to extract hierarchical visual features. Real-world applications are highlighted — including object detection, facial recognition, autonomous vehicles, and medical image analysis — to demonstrate how CNNs revolutionize visual data understanding. The presentation also touches on recent advancements such as transfer learning, pre-trained models (like VGG, ResNet), and frameworks used for implementation (TensorFlow, PyTorch). By the end, the audience will have a clear understanding of how CNNs mimic human visual perception to perform complex image recognition tasks with high accuracy.

Embed presentation

Download to read offline
1 / 116
2 / 116
3 / 116
4 / 116
5 / 116
6 / 116
7 / 116
8 / 116
9 / 116
10 / 116
11 / 116
12 / 116
13 / 116
14 / 116
15 / 116
16 / 116
17 / 116
18 / 116
19 / 116
20 / 116
21 / 116
22 / 116
23 / 116
24 / 116
25 / 116
26 / 116
27 / 116
28 / 116
29 / 116
30 / 116
31 / 116
32 / 116
33 / 116
34 / 116
35 / 116
36 / 116
37 / 116
38 / 116
39 / 116
40 / 116
41 / 116
42 / 116
43 / 116
44 / 116
45 / 116
46 / 116
47 / 116
48 / 116
49 / 116
50 / 116
51 / 116
52 / 116
53 / 116
54 / 116
55 / 116
56 / 116
57 / 116
58 / 116
59 / 116
60 / 116
61 / 116
62 / 116
63 / 116
64 / 116
65 / 116
66 / 116
67 / 116
68 / 116
69 / 116
70 / 116
71 / 116
72 / 116
73 / 116
74 / 116
75 / 116
76 / 116
77 / 116
78 / 116
79 / 116
80 / 116
81 / 116
82 / 116
83 / 116
84 / 116
85 / 116
86 / 116
87 / 116
88 / 116
89 / 116
90 / 116
91 / 116
92 / 116
93 / 116
94 / 116
95 / 116
96 / 116
97 / 116
98 / 116
99 / 116
100 / 116
101 / 116
102 / 116
103 / 116
104 / 116
105 / 116
106 / 116
107 / 116
108 / 116
109 / 116
110 / 116
111 / 116
112 / 116
113 / 116
114 / 116
115 / 116
116 / 116
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars 6.S094: Deep Learning for Self-Driving Cars Learning to Drive: Convolutional Neural Networks and End-to-End Learning of the Full Driving Tasks cars.mit.edu
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Administrative • Website: cars.mit.edu • Contact Email: deepcars@mit.edu • Required: • Create an account on the website. • Follow the tutorial for each of the 2 projects. • Recommended: • Ask questions • Win competition! • Office hours: Friday, 5-7pm (more info coming soon)
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Administrative • Website: cars.mit.edu • Contact Email: deepcars@mit.edu • Required: • Create an account on the website. • Follow the tutorial for each of the 2 projects. • Recommended: • Ask questions • Win competition!
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Schedule
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars DeepTraffic Leaderboard
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Illustrative Case Study: Traffic Light Detection
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars DeepTesla: End-to-End Learning from Human and Autopilot Driving (in ConvnetJS)
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars DeepTesla: End-to-End Learning from Human and Autopilot Driving (in TensorFlow)
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Supervised Learning Unsupervised Learning Semi-Supervised Learning Reinforcement Learning Standard supervised learning pipeline: Computer Vision is Machine Learning References: [81] Computer Vision
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Images are Numbers References: [89] • Regression: The output variable takes continuous values • Classification: The output variable takes class labels • Underneath it may still produce continuous values such as probability of belonging to a particular class.
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Computer Vision is Hard References: [66, 69, 89]
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Image Classification Pipeline References: [81, 89]
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Famous Computer Vision Datasets References: [90, 91, 92, 93] MNIST: handwritten digits ImageNet: WordNet hierarchy CIFAR-10(0): tiny images Places: natural scenes
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Let’s Build an Image Classifier for CIFAR-10 References: [89, 91]
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Let’s Build an Image Classifier for CIFAR-10 References: [89, 91] Accuracy Random: 10% Our image-diff (with L1): 38.6% Our image-diff (with L2): 35.4%
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars K-Nearest Neighbors: Generalizing the Image-Diff Classifier References: [89] Tuning (hyper)parameters:
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars K-Nearest Neighbors: Generalizing the Image-Diff Classifier References: [89, 94] Accuracy Random: 10% Training and testing on the same data: 35.4% 7-Nearest Neighbors: ~30% Human: ~94% … Convolutional Neural Networks: ~95%
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Reminder: Weighing the Evidence References: [78] Evidence Decisions
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Reminder: Classify and Image of a Number References: [80] Input: (28x28) Network:
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Reminder: “Learning” is Optimization of a Function References: [63, 80] Ground truth for “6”: “Loss” function:
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Convolutional Neural Networks References: [95] Regular neural network (fully connected): Convolutional neural network: Each layer takes a 3d volume, produces 3d volume with some smooth function that may or may not have parameters.
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Convolutional Neural Networks: Layers • INPUT [32x32x3] will hold the raw pixel values of the image, in this case an image of width 32, height 32, and with three color channels R,G,B. • CONV layer will compute the output of neurons that are connected to local regions in the input, each computing a dot product between their weights and a small region they are connected to in the input volume. This may result in volume such as [32x32x12] if we decided to use 12 filters. • RELU layer will apply an elementwise activation function, such as the max(0,x) thresholding at zero. This leaves the size of the volume unchanged ([32x32x12]). • POOL layer will perform a downsampling operation along the spatial dimensions (width, height), resulting in volume such as [16x16x12]. • FC (i.e. fully-connected) layer will compute the class scores, resulting in volume of size [1x1x10], where each of the 10 numbers correspond to a class score, such as among the 10 categories of CIFAR-10. As with ordinary Neural Networks and as the name implies, each neuron in this layer will be connected to all the numbers in the previous volume. References: [95]
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Dealing with Images: Local Connectivity Same neuron. Just more focused (narrow “receptive field”). The parameters on a each filter are spatially “shared” (if a feature is useful in one place, it’s useful elsewhere) References: [95]
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars ConvNets: Spatial Arrangement of Output Volume • Depth: number of filters • Stride: filter step size (when we “slide” it) • Padding: zero-pad the input References: [95]
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars References: [95]
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars ConvNets: Pooling References: [95]
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Computer Vision: Object Recognition / Classification References: [4]
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Original Ground Truth FCN-8 Computer Vision: Segmentation References: [96]
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Computer Vision: Object Detection References: [97]
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars How Can Convolutional Neural Networks Help Us Drive?
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Driving: The Numbers (in United States, in 2014) • All drivers: 10,658 miles (29.2 miles per day) • Rural drivers: 12,264 miles • Urban drivers: 9,709 miles • Fatal crashes: 29,989 • All fatalities: 32,675 • Car occupants: 12,507 • SUV occupants: 8,320 • Pedestrians: 4,884 • Motorcycle: 4,295 • Bicyclists: 720 • Large trucks: 587 Miles Fatalities
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Cars We Drive
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Human at the Center of Automation: The Way to Full Autonomy Includes the Human Ford F150 Tesla Model S Google Self-Driving Car Fully Human Controlled Fully Machine Controlled
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Human at the Center of Automation: The Way to Full Autonomy Includes the Human • Emergency • Automatic emergency breaking (AEB) • Warnings • Lane departure warning (LDW) • Forward collision warning (FCW) • Blind spot detection • Longitudinal • Adaptive cruise control (ACC) • Lateral • Lane keep assist (LKA) • Automatic steering • Control and Planning • Automatic lane change • Automatic parking Tesla Autopilot
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Distracted Humans • Injuries and fatalities: 3,179 people were killed and 431,000 were injured in motor vehicle crashes involving distracted drivers (in 2014) • Texts: 169.3 billion text messages were sent in the US every month. (as of December 2014) • Eye off road: 5 seconds is the average time your eyes are off the road while texting. When traveling at 55mph, that's enough time to cover the length of a football field blindfolded. What is distracted driving? • Texting • Using a smartphone • Eating and drinking • Talking to passengers • Grooming • Reading, including maps • Using a navigation system • Watching a video • Adjusting a radio
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars 4 D’s of Being Human: Drunk, Drugged, Distracted, Drowsy • Drunk Driving: In 2014, 31 percent of traffic fatalities involved a drunk driver. • Drugged Driving: 23% of night-time drivers tested positive for illegal, prescription or over-the-counter medications. • Distracted Driving: In 2014, 3,179 people (10 percent of overall traffic fatalities) were killed in crashes involving distracted drivers. • Drowsy Driving: In 2014, nearly three percent of all traffic fatalities involved a drowsy driver, and at least 846 people were killed in crashes involving a drowsy driver.
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars In Context: Traffic Fatalities Total miles driven in U.S. in 2014: 3,000,000,000,000 (3 million million) Fatalities: 32,675 (1 in 90 million) Tesla Autopilot miles driven since October 2015: 300,000,000 (300 million) (as of December 2016) Fatalities: 1
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars In Context: Traffic Fatalities Total miles driven in U.S. in 2014: 3,000,000,000,000 (3 million million) Fatalities: 32,675 (1 in 90 million) Tesla Autopilot miles driven since October 2015: 300,000,000 (300 million) Fatalities: 1
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars In Context: Traffic Fatalities Total miles driven in U.S. in 2014: 3,000,000,000,000 (3 million million) Fatalities: 32,675 (1 in 90 million) Tesla Autopilot miles driven since October 2015: 300,000,000 (300 million) Fatalities: 1 We (increasingly) understand this We do not understand this (yet) We need A LOT of real-world semi-autonomous driving data! Computer Vision + Machine Learning + Big Data = Understanding
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars The Data Teslas instrumented: 17 Hours of data: 5,000+ hours Distance traveled: 70,000+ miles
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars The Data
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Camera and Lens Selection Fisheye: Capture full range of head, body movement inside vehicle. 2.8-12mm Focal Length: “Zoom” on the face without obstructing the driver’s view. Logitech C920: On-board H264 Compression Case for C-Mount Lens: Flexibility in lens selection
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Semi-Autonomous Vehicle Components External 1. Radar 2. Visible-light camera 3. LIDAR 4. Infrared camera 5. Stereo vision 6. GPS/IMU 7. CAN 8. Audio Internal 1. Visible-light camera 2. Infrared camera 3. Audio
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Self-Driving Car Tasks • Localization and Mapping: Where am I? • Scene Understanding: Where is everyone else? • Movement Planning: How do I get from A to B? • Driver State: What’s the driver up to?
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Self-Driving Car Tasks • Localization and Mapping: Where am I? • Scene Understanding: Where is everyone else? • Movement Planning: How do I get from A to B? • Driver State: What’s the driver up to?
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Visual Odometry • 6-DOF: freed of movement • Changes in position: • Forward/backward: surge • Left/right: sway • Up/down: heave • Orientation: • Pitch, Yaw, Roll • Source: • Monocular: I moved 1 unit • Stereo: I moved 1 meter • Mono = Stereo for far away objects • PS: For tiny robots everything is “far away” relative to inter-camera distance
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars SLAM: Simultaneous Localization and Mapping What works: SIFT and optical flow References: [98, 99]
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Visual Odometry in Parts • (Stereo) Undistortion, Rectification • (Stereo) Disparity Map Computation • Feature Detection (e.g., SIFT, FAST) • Feature Tracking (e.g., KLT: Kanade-Lucas-Tomasi) • Trajectory Estimation • Use rigid parts of the scene (requires outlier/inlier detection) • For mono, need more info* like camera orientation and height of off the ground * Kitt, Bernd Manfred, et al. "Monocular visual odometry using a planar road model to solve scale ambiguity." (2011).
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars End-to-End Visual Odometry Konda, Kishore, and Roland Memisevic. "Learning visual odometry with a convolutional network." International Conference on Computer Vision Theory and Applications. 2015.
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Self-Driving Car Tasks • Localization and Mapping: Where am I? • Scene Understanding: Where is everyone else? • Movement Planning: How do I get from A to B? • Driver State: What’s the driver up to?
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Object Detection • Past approaches: cascades classifiers (Haar-like features) • Where deep learning can help: recognition, classification, detection
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Full Driving Scene Segmentation Fully Convolutional Network implementation: https://github.com/tkuanlun350/Tensorflow-SegNet
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Road Texture and Condition from Audio (with Recurrent Neural Networks)
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Self-Driving Car Tasks • Localization and Mapping: Where am I? • Scene Understanding: Where is everyone else? • Movement Planning: How do I get from A to B? • Driver State: What’s the driver up to?
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars • Previous approaches: optimization-based control • Where deep learning can help: reinforcement learning Deep Reinforcement Learning implementation: https://github.com/nivwusquorum/tensorflow-deepq
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Self-Driving Car Tasks • Localization: Where am I? • Object detection: Where is everyone else? • Movement planning: How do I get from A to B? • Driver state: What’s the driver up to?
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Drive State Detection: A Multi-Resolutional View Gaze Classification Blink Rate Blink Duration Head Pose Eye Pose Pupil Diameter Micro Saccades Increasing level of detection resolution and difficulty Body Pose Blink Dynamics Micro Glances Cognitive Load Drowsiness
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Gaze Region and Autopilot State
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Driver Emotion
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars If Driving is a Conversation, this is an End-to-End Natural Language Generation 1. Natural language processing to enable it to communicate successfully 2. Knowledge representation to store information provided before or during the interrogation 3. Automated reasoning to use the stored information to answer questions and to draw new conclusions Turing Test: Can a computer be mistaken for a human more than 30% of the time?
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Autonomous Driving: End-to-End Magic Happens
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Stairway to Automation Ford F150 Tesla Model S Google Self-Driving Car Training Dataset Testing Dataset
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Autonomous Driving: End-to-End
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Autonomous Driving: End-to-End
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Autonomous Driving: End-to-End • 9 layers • 1 normalization layer • 5 convolutional layers • 3 fully connected layers • 27 million connections • 250 thousand parameters
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars End-to-End Driving with ConvnetJS Tutorial on http://cars.mit.edu/deeptesla
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars End-to-end Steering • By the end of this lecture, you’ll be able to train a model that can steer a vehicle • Our input to our network will be a single image of the forward roadway from a Tesla • Our output will be a steering wheel value between -20 and 20
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Creating the Dataset • We recorded and extracted 10 video clips of highway driving from a Tesla • The wheel value was extracted from the in-vehicle CAN • We cropped/extracted a window from each video frame and provide a CSV linking the window to a wheel value
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Lighting and Road Conditions
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars ConvNetJS Overview • ConvNetJS is a Javascript implementation for using and training neural networks within the browser • It supports simple networks with several different layer types and training algorithms • Constructing and training a network can be performed in very few lines of code, great for demonstrations
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars ConvNetJS – Neural Network Representation • The network is represented by a single Javascript object which contains a list of layers • Each layer contains a plain array of weights (w), the activation/activation gradients of the last forward pass, as well as the shape and layer type
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Layer Types • ConvNetJS implements several different layer types: convolutional, pooling, fully-connected, local contrast normalization, and loss layers • There are three available output types: regression, softmax, and SVM
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars ConvNetJS – Training Overview • To train a network, you first must initialize a “Trainer” object • var trainer = new convnetjs.SGDTrainer(net, { method: ‘adadelta’, batch_size: 1, l2_decay: 0.0001}); • There are three training algorithms available: SGD, Adadelta, and Adagrad. • Training is performed by manually calling trainer.train(input_volume, expected_output) • Returns an object containing timing and loss function information
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars DeepTesla Overview
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Model Metrics
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Network Designer
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Training Interaction
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Layer Visualization
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Input Layer
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Convolutional Layer Visualization
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Video Visualization
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Information Bar
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Input Box
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Barcodes • 17 bit, sign-magnitude • Encoded into actual video • 0 = black, 1 = white • Frame on top, wheel on bottom (divided by two)
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Image Batches • Each image loaded of the network contains an entire batch • There is one image per row, and 250 rows in total • These images are reassembled into volumes upon download
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Training Explanation • One web worker used for loading examples • Each batch of training images is one large image with each row as a single training example • After an image finishes loading asynchronously, sends the training examples to another worker • One web worker used for training network • Train on each image and push the network/outputs to visualization worker • One web worker used for visualization • For a specified training example interval, blit the activation/gradient output of each training example onto a canvas • Each web worker behaves as a single thread, and we use message passing to communicate state between the workers
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars ConvNetJS Evaluation - Video Explanation • The videos are encoded as 1280x820 in H264/MKV with 17 bit sign-magnitude barcodes • The video main video frame is stored in the box (0, 1280, 0, 720) • The frame barcode is in box (1144, 720, 1280, 770) • The wheel value barcode is in box (1144, 770, 1280, 820)
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars ConvNetJS Evaluation - Creating the Video • Each epoch is synchronized to 30 fps • We extract the wheel value from the CAN data and synchronize each message to a frame (both the frame and the CAN message are timestamped) • Using OpenCV, we process the data • Generate a bar code for the frame containing the wheel data • Crop the image portion used for training • Create single images containing batches of training data • The epochs and associated data are copied to our web server which serves
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars ConvNetJS Evaluation - Playing the Video • To be able to use the video in the neural network, we need to do some preprocessing • First, we have a hidden video element and rely on modern HTML5 video implementations • When the user requests the video to play, we begin tracking each redraw of the page • With each redraw, we grab the currently rendered video frame, extract the RGBA values and blit them to two different canvases: one canvas, which the user sees, and another canvas which is hidden and only contains a cropped portion of the frame (the part we will use for the neural network)
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars ConvNetJS Evaluation - Playing the Video • Next, we read the image data from the hidden canvas and shape it into a ConvNetJS volume • For each image we first create a volume: • var image_vol = new convnetjs.Vol(x_size, y_size, depth, default_value) • Next, we extract each pixel from the canvas and set the equivalent voxel (volume pixel) to the value (skipping the alpha value) • We can also extract the expected steer value by parsing the barcode (a 17 bit, signed-magnitude barcode, where white = 1, black = 0)
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars ConvNetJS Evaluation - Forward Pass • Now we can use our extracted volume in the forward pass by calling net.forward(our_volume) • The predicted value is stored in the output neuron: • var prediction = net.forward(vol); • var raw_regression_value = prediction.w[0]; • Because we min-max normalized our inputs while training the network, we need to transform our outputs – this is just the reverse transformation we performed on input: • Wheel value = (raw_regression_value * total_wheel_range) – wheel_min • We visualize the predicted and actual steering wheel values and calculate the error
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars End-to-End Driving with ConvnetJS Tutorial on http://cars.mit.edu/deeptesla
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars End-to-End Driving with TensorFlow Available on http://github.com/lexfridman/deeptesla
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Build the Model: Input and Output def weight_variable(shape): initial = tf.truncated_normal(shape, stddev=0.1) return tf.Variable(initial) def bias_variable(shape): initial = tf.constant(0.1, shape=shape) return tf.Variable(initial) def conv2d(x, W, stride): return tf.nn.conv2d(x, W, strides=[1, stride, stride, 1], padding='VALID') x = tf.placeholder(tf.float32, shape=[None, 66, 200, 3]) y_ = tf.placeholder(tf.float32, shape=[None, 1]) x_image = x
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Build the Model: Convolutional Layers #first convolutional layer W_conv1 = weight_variable([5, 5, 3, 24]) b_conv1 = bias_variable([24]) h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1, 2) + b_conv1) #second convolutional layer W_conv2 = weight_variable([5, 5, 24, 36]) b_conv2 = bias_variable([36]) h_conv2 = tf.nn.relu(conv2d(h_conv1, W_conv2, 2) + b_conv2) #third convolutional layer W_conv3 = weight_variable([5, 5, 36, 48]) b_conv3 = bias_variable([48]) h_conv3 = tf.nn.relu(conv2d(h_conv2, W_conv3, 2) + b_conv3) #fourth convolutional layer W_conv4 = weight_variable([3, 3, 48, 64]) b_conv4 = bias_variable([64]) h_conv4 = tf.nn.relu(conv2d(h_conv3, W_conv4, 1) + b_conv4) #fifth convolutional layer W_conv5 = weight_variable([3, 3, 64, 64]) b_conv5 = bias_variable([64]) h_conv5 = tf.nn.relu(conv2d(h_conv4, W_conv5, 1) + b_conv5)
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Build the Model: Fully Connected Layers # fully connected layer 1 W_fc1 = weight_variable([1152, 1164]) b_fc1 = bias_variable([1164]) h_conv5_flat = tf.reshape(h_conv5, [-1, 1152]) h_fc1 = tf.nn.relu(tf.matmul(h_conv5_flat, W_fc1) + b_fc1) keep_prob = tf.placeholder(tf.float32) h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob) # fully connected layer 2 W_fc2 = weight_variable([1164, 100]) b_fc2 = bias_variable([100]) h_fc2 = tf.nn.relu(tf.matmul(h_fc1_drop, W_fc2) + b_fc2) h_fc2_drop = tf.nn.dropout(h_fc2, keep_prob) # fully connected layer 3 W_fc3 = weight_variable([100, 50]) b_fc3 = bias_variable([50]) h_fc3 = tf.nn.relu(tf.matmul(h_fc2_drop, W_fc3) + b_fc3) h_fc3_drop = tf.nn.dropout(h_fc3, keep_prob) # fully connected layer 4 W_fc4 = weight_variable([50, 10]) b_fc4 = bias_variable([10]) h_fc4 = tf.nn.relu(tf.matmul(h_fc3_drop, W_fc4) + b_fc4) h_fc4_drop = tf.nn.dropout(h_fc4, keep_prob) #Output W_fc5 = weight_variable([10, 1]) b_fc5 = bias_variable([1]) y = tf.mul(tf.atan(tf.matmul(h_fc4_drop, W_fc5) + b_fc5), 2)
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Train the Model sess = tf.InteractiveSession() loss = tf.reduce_mean(tf.square(tf.sub(model.y_, model.y))) train_step = tf.train.AdamOptimizer(1e-4).minimize(loss) sess.run(tf.initialize_all_variables()) saver = tf.train.Saver() for i in range(int(driving_data.num_images * 0.3)): xs, ys = driving_data.LoadTrainBatch(100) train_step.run(feed_dict={model.x: xs, model.y_: ys, model.keep_prob: 0.8}) if i % 10 == 0: xs, ys = driving_data.LoadValBatch(100) print("step %d, val loss %g"%(i, loss.eval(feed_dict={ model.x:xs, model.y_: ys, model.keep_prob: 1.0}))) if i % 100 == 0: if not os.path.exists(LOGDIR): os.makedirs(LOGDIR) checkpoint_path = os.path.join(LOGDIR, "model.ckpt") filename = saver.save(sess, checkpoint_path) print("Model saved in file: %s" % filename)
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Run the Model import tensorflow as tf import scipy.misc import model import cv2 sess = tf.InteractiveSession() saver = tf.train.Saver() saver.restore(sess, "save/model.ckpt") img = cv2.imread('steering_wheel_image.jpg',0) rows,cols = img.shape smoothed_angle = 0 cap = cv2.VideoCapture(0) while(cv2.waitKey(10) != ord('q')): ret, frame = cap.read() image = scipy.misc.imresize(frame, [66, 200]) / 255.0 degrees = model.y.eval(feed_dict={model.x: [image], model.keep_prob: 1.0})[0][0] * 180 / scipy.pi cv2.imshow('frame', frame) smoothed_angle += 0.2 * pow(abs((degrees - smoothed_angle)), 2.0 / 3.0) * (degrees - smoothed_angle) / abs(degrees - smoothed_angle) M = cv2.getRotationMatrix2D((cols/2,rows/2),-smoothed_angle,1) dst = cv2.warpAffine(img,M,(cols,rows)) cv2.imshow("steering wheel", dst) cap.release() cv2.destroyAllWindows()
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars End-to-End Driving with TensorFlow Available on http://github.com/lexfridman/deeptesla
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars TrafficLight Classification with TensorFlow We will be implementing a simple traffic light classifier, with 3 classes (red, green, yellow)
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Parameters • Max epochs: the number of times the neural network will see all training examples • Input_img_x/y: the size we will use for inputs into the the network • Batch size: # of examples the neural network will see before making a gradient step
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Helper Functions We use some helper functions to make adding layers easier/more consistent
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Model Input/Output • We specify our input and output types in the same lines to make sure they agree with our idea of the network • Our input is an image of sized 32x32x3 (RGB channels) • Our output consists of 3 neurons, representing the probability of each class
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Convolutional Layer • Here we specify our first convolutional layer using our helper function • W_conv1 – a 4D tensor representing the weights [filter_x, filter_y, previous layer neurons, # of filters] • b_conv1 – our simple addition variable • h_conv1 – our actual layer/activation
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Pooling Layer
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Flattening Pool Layer We calculate the total number of neurons needed in our first fully-connected layer by multiplying all the dimensions of the pool layer shape
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Output Layer
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Loss and Optimizer • Our loss function performs softmax and then computes cross-entropy • We use the AdamOptimizer and specify a learning rate
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Saver Object
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Loading Images • Iterate over each image, resize to 32x32 • Create a one hot encoding of our class • Shuffle the entire dataset
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Splitting the Dataset • Split our data set into train and test • We truncate our sets to a multiple of batch size (all batches have to be the same size)
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Training Loop • Iterate over each batch and train on it • (we assume training examples are a multiple of the batch size)
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Best Model • We evaluate the loss on all of our training examples and test examples • If the validation loss is lower than the lowest loss, we save our model
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars Expected Output
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars TrafficLight Classification with TensorFlow We will be implementing a simple traffic light classifier, with 3 classes (red, green, yellow)
Lex Fridman: fridman@mit.edu Website: cars.mit.edu January 2017 Course 6.S094: Deep Learning for Self-Driving Cars References All references cited in this presentation are listed in the following Google Sheets file: https://goo.gl/9Xhp2t

More Related Content

PPTX
Computer Vision And Convolutional Neural Networks (CNNs)
byadityasiwach20
 
PPTX
ObjectDetection.pptx
byRitikPabbaraju2
 
PDF
Deep learning state_of_the_art- Autonomous Driving
byAlok Jain
 
PDF
Computer vision for transportation
byWanjin Yu
 
PPTX
Self-Driving Cars With Convolutional Neural Networks (CNN.pptx
byssuserf79e761
 
PPTX
Deep Learning Algorithm Using Virtual Environment Data For Self-Driving Car
bysushilkumar1236
 
PDF
Developing and Deploying Deep Learning Based Computer Vision Systems - Alka N...
byCodeOps Technologies LLP
 
PDF
Deep Learning and the technology behind Self-Driving Cars
byLucas García, PhD
 
Computer Vision And Convolutional Neural Networks (CNNs)
byadityasiwach20
 
ObjectDetection.pptx
byRitikPabbaraju2
 
Deep learning state_of_the_art- Autonomous Driving
byAlok Jain
 
Computer vision for transportation
byWanjin Yu
 
Self-Driving Cars With Convolutional Neural Networks (CNN.pptx
byssuserf79e761
 
Deep Learning Algorithm Using Virtual Environment Data For Self-Driving Car
bysushilkumar1236
 
Developing and Deploying Deep Learning Based Computer Vision Systems - Alka N...
byCodeOps Technologies LLP
 
Deep Learning and the technology behind Self-Driving Cars
byLucas García, PhD
 

Similar to Convolutional Neural Networks(CNN) and Computer Vision

PDF
Deep Learning State of the Art (2019) - MIT by Lex Fridman
byPeerasak C.
 
PDF
Self-Driving Car to Drive Autonomously using Image Processing and Deep Learning
byIRJET Journal
 
PPTX
“ADAS in Action (POC Autonomous Driving Vehicle Presentation)”
byGlobalLogic Ukraine
 
PDF
Forrest Iandola: My Adventures in Artificial Intelligence and Entrepreneurship
byForrest Iandola
 
PDF
Opening Keynote at GTC 2015: Leaps in Visual Computing
byNVIDIA
 
PPTX
Dov Nimratz, Roman Chobik "Embedded artificial intelligence"
byLviv Startup Club
 
PDF
Techniques and Challenges in Autonomous Driving
byYu Huang
 
PPTX
IntroComputerVision23.pptx
byAneesAbbasi14
 
PDF
BMW：Application and Deployment of Deep Learning in Autonomous Driving.pdf
bykeithzhangding
 
PDF
Advance Self Driving Car Using Machine Learning
byAssociate Professor in VSB Coimbatore
 
PDF
An Analysis of Various Deep Learning Algorithms for Image Processing
byvivatechijri
 
PDF
IRJET - IoT based Self Driving Car
byIRJET Journal
 
PDF
IRJET- Self Driving RC Car using Behavioral Cloning
byIRJET Journal
 
PDF
Imaging automotive 2015 addfor v002
byEnrico Busto
 
PDF
Imaging automotive 2015 addfor v002
byEnrico Busto
 
PDF
Object Detection Beyond Mask R-CNN and RetinaNet II
byWanjin Yu
 
PPTX
Neurons On Wheels - Implementation
byNECST Lab @ Politecnico di Milano
 
PPTX
1_Intro2ssssssssssssssssssssssssssssss2.pptx
bylarturo
 
PPTX
2nd DL Meetup @ Dublin - Irene
byZihui Li
 
PDF
Material
byMohamedTarek424
 
Deep Learning State of the Art (2019) - MIT by Lex Fridman
byPeerasak C.
 
Self-Driving Car to Drive Autonomously using Image Processing and Deep Learning
byIRJET Journal
 
“ADAS in Action (POC Autonomous Driving Vehicle Presentation)”
byGlobalLogic Ukraine
 
Forrest Iandola: My Adventures in Artificial Intelligence and Entrepreneurship
byForrest Iandola
 
Opening Keynote at GTC 2015: Leaps in Visual Computing
byNVIDIA
 
Dov Nimratz, Roman Chobik "Embedded artificial intelligence"
byLviv Startup Club
 
Techniques and Challenges in Autonomous Driving
byYu Huang
 
IntroComputerVision23.pptx
byAneesAbbasi14
 
BMW：Application and Deployment of Deep Learning in Autonomous Driving.pdf
bykeithzhangding
 
Advance Self Driving Car Using Machine Learning
byAssociate Professor in VSB Coimbatore
 
An Analysis of Various Deep Learning Algorithms for Image Processing
byvivatechijri
 
IRJET - IoT based Self Driving Car
byIRJET Journal
 
IRJET- Self Driving RC Car using Behavioral Cloning
byIRJET Journal
 
Imaging automotive 2015 addfor v002
byEnrico Busto
 
Imaging automotive 2015 addfor v002
byEnrico Busto
 
Object Detection Beyond Mask R-CNN and RetinaNet II
byWanjin Yu
 
Neurons On Wheels - Implementation
byNECST Lab @ Politecnico di Milano
 
1_Intro2ssssssssssssssssssssssssssssss2.pptx
bylarturo
 
2nd DL Meetup @ Dublin - Irene
byZihui Li
 
Material
byMohamedTarek424
 

Recently uploaded

PDF
NYC ACE 12-Nov-2025-Combined Presentation.pdf
byAUGNYC
 
PDF
AI/ML Infra Meetup | SkyPilot: Open-source System to Scale AI across Clusters...
byAlluxio, Inc.
 
PDF
Connecting Strategy to Delivery Teams Using OnePlan.pdf
byOnePlan Solutions
 
PDF
AI/ML Infra Meetup | Unlock the Future of Generative AI: TorchTitan's Latest ...
byAlluxio, Inc.
 
PDF
A new Vision of Software Sustainability and its Engineering: Reflections and ...
byPatricia Lago
 
PDF
Startup Core Architecture: A Simple Practical Guide
byShiv Technolabs Pvt. Ltd.
 
PDF
The Evolution of Project Portfolio Management - What Modern PMOs Are Doing Di...
byOnePlan Solutions
 
PDF
How Integrating Copilot and AI Agents in Microsoft Dynamics 365
byTechWize
 
PPTX
Top Benefits of Hiring a Mobile App Development Company in India
byDevherds software solution
 
DOCX
Open Source BI Tableau Alternatives You Can Trust.docx
byVarsha Nayak
 
PDF
How to Make Your AI Reliable: Comprehensive Guide to Testing AI Applications ...
byQATestLab Quality is a rule
 
PDF
Quantum TechDocs - Technology & Compliance.pdf
byietmsoftware1
 
PDF
Hyperon Chain – Smart Contract Security Audit Report by EtherAuthority
byEtherAuthority1
 
PDF
6 Hotel Booking Trends You Can’t Ignore in 2025.pdf
byHotelogix
 
PDF
A Complete Guide to Salesforce for Education.pdf
byVALiNTRY360
 
PDF
vMix Overview & Key Features Easily Produce, Record and Live Stream
byIGM Soraing
 
PDF
Combinatorial Interview Problems with Backtracking Solutions - From Imperativ...
byPhilip Schwarz
 
PPTX
Working of Minimax Algorithm with Alpha-Beta Pruning.pptx
byMuhammadHassanArshad2
 
PDF
DSD-INT 2025 The Singapore Regional Model in Delft3D FM - Zijl
byDeltares
 
PPTX
Short Film Script Development Process HARMONIA
byzdolezalmedia
 
NYC ACE 12-Nov-2025-Combined Presentation.pdf
byAUGNYC
 
AI/ML Infra Meetup | SkyPilot: Open-source System to Scale AI across Clusters...
byAlluxio, Inc.
 
Connecting Strategy to Delivery Teams Using OnePlan.pdf
byOnePlan Solutions
 
AI/ML Infra Meetup | Unlock the Future of Generative AI: TorchTitan's Latest ...
byAlluxio, Inc.
 
A new Vision of Software Sustainability and its Engineering: Reflections and ...
byPatricia Lago
 
Startup Core Architecture: A Simple Practical Guide
byShiv Technolabs Pvt. Ltd.
 
The Evolution of Project Portfolio Management - What Modern PMOs Are Doing Di...
byOnePlan Solutions
 
How Integrating Copilot and AI Agents in Microsoft Dynamics 365
byTechWize
 
Top Benefits of Hiring a Mobile App Development Company in India
byDevherds software solution
 
Open Source BI Tableau Alternatives You Can Trust.docx
byVarsha Nayak
 
How to Make Your AI Reliable: Comprehensive Guide to Testing AI Applications ...
byQATestLab Quality is a rule
 
Quantum TechDocs - Technology & Compliance.pdf
byietmsoftware1
 
Hyperon Chain – Smart Contract Security Audit Report by EtherAuthority
byEtherAuthority1
 
6 Hotel Booking Trends You Can’t Ignore in 2025.pdf
byHotelogix
 
A Complete Guide to Salesforce for Education.pdf
byVALiNTRY360
 
vMix Overview & Key Features Easily Produce, Record and Live Stream
byIGM Soraing
 
Combinatorial Interview Problems with Backtracking Solutions - From Imperativ...
byPhilip Schwarz
 
Working of Minimax Algorithm with Alpha-Beta Pruning.pptx
byMuhammadHassanArshad2
 
DSD-INT 2025 The Singapore Regional Model in Delft3D FM - Zijl
byDeltares
 
Short Film Script Development Process HARMONIA
byzdolezalmedia
 

Convolutional Neural Networks(CNN) and Computer Vision