Deep Learning for Image

A practioners' perspective on doing Deep Learning for Image

Transcript

1. Deep Learning for Image A practitioner’s perspective Amit Kapoor amitkaps.com Bargava Subramanian bargava.com
2. Practitioners Amit Bargava
3. Outline for today... 1. Why deep learning now? 2. How to adopt a practical approach? a. Learning b. Data c. Tools & Deploy 3. Where do you go from here?
4. Outline for today... 1. Why deep learning now? 2. How to adopt a practical approach? a. Learning b. Data c. Tools & Deploy 3. Where do you go from here?
5. Classical Programming Paradigm Input → ? → Output
6. Classical Programming Paradigm Input → ? → Output user types the text 4 Update database 4
7. Task: Write the function Input → f(x) → Output Write the function user types the text 4 Update database 4
8. Challenge: Robust Functions Input → f(x) → Output Write the function user types the text 4. Update database 4 Challenge Test to ensure it is robust for all possible inputs
9. Learning Paradigm Input → ? → Output Learn the function user writes the text Update database 4
10. Task: Create Features & Learn Input → Feature → g(x) → Output Create features user types the text Update database 4 Learn the Function
11. Input → Feature → g(x) → Output Challenge: Hand-crafted features Create features user types the text Update database 4 Learn the Function Challenge How do I hand-craft the right set of features to learn the function
12. Input → f(x) → Output Input → Features → g(x) → Output learn Two Paradigms Create Create Learning Paradigm Programming Paradigm
13. - Structure of Data (tabular, text, image, video, sound) - Amount of Data (none, small, medium, large) - Knowledge of Domain (limited, expert) When to use which paradigm?
14. Input → Features → g(x) → Output learn Learning Paradigm Create Traditional Machine Learning
15. Input → Features → g(x) → Output Input → Features → h(x) → Output learn Deep Learning Create Deep Learning Traditional Machine Learning learn
16. What is deep about it? user types the text Update database 4 Layer 1 Layer 2 Layer 3 Layer 4 Learning Higher Order Represenastions
17. What is deep about it? Source: Deep Learning by Francois Chollet
18. - Access to more Data - Faster Compute (using GPUs) - Clever Algorithmic choices Why now?
19. Open Discussion on use Cases - Tabular - Text - Image - Video - Speech
20. Outline for today... 1. Why deep learning now? 2. How to adopt a practical approach? a. Learning b. Data c. Tools & Deploy 3. Where do you go from here?
21. Image: Logo Detection Industry Ad Tech Objective User engagement Outcome Targeted ads on digital media
22. Image: Traffic Sign Detection Industry Self-driving cars Objective Traffic Sign Adherence Outcome Traffic Sign in native language
23. Architecture Design Spatial structure Weights/Overfitting
24. Model: Convolutional Neural Network Key ideas for image 1. Local Receptive Fields 2. Shared Weights 3. Sub-sampling
25. Local Receptive Fields: Convolution Input Image Conv Kernel Output
26. Shared Weights: localized feature maps - One feature map detects a single kind of localized feature - Use several feature maps
27. Sub-sampling: Max Pooling
28. CNN: Architecture
29. Outline for today... 1. Why deep learning now? 2. How to adopt a practical approach? a. Learning b. Data c. Tools & Deploy 3. Where do you go from here?
30. Data: Input Structure - Varied input sizes - Color images - Around 20k images
31. Input: Pre-processing - Zero-centered X = X - np.mean(X, axis = 0) - Normalization X = X / np.std(X, axis = 0)
32. In the wild : CNN from scratch (1/2) - Define architecture - Smart weight initialization (e.g. Xavier)
33. In the wild : CNN from scratch (2/2) Could we do better?
34. First model : Transfer Learning Pre-trained model - Model built on a large dataset (eg: ImageNet) - Most libraries have model zoo - architecture with final trained weights
35. Pre-trained model : vgg16 First model to surpass human-level performance on ImageNet
36. Transfer Learning: Practicalities Less Data More Data Same Domain Retrain last classifier layer Fine tune last few layers Different Domain TROUBLE !! Fine tune a number of layers
37. Less Data More Data Same Domain Retrain last classifier layer Fine tune last few layers Different Domain TROUBLE !! Fine tune a number of layers Pre-trained models: Initial results We started here! Using pre-trained models - achieved 88% accuracy. < 10 min train time
38. Client needed 95% accuracy Needed more data !
39. Outline for today... 1. Why deep learning now? 2. How to adopt a practical approach? a. Learning b. Data c. Tools & Deploy 3. Where do you go from here?
40. More data: Approaches - Augmentation - Generation
41. Data: Augmentation - Horizontal/Vertical Flips - Scale - Random cropping - Jitter Any combination(s) of: - Translation - Rotation - Stretching - Shearing - Lens distortion
42. Generation: Why? - Need images in different conditions e.g. snow, rain, fog - Models and compute better than manually coding many possibilities
43. Data: Generation - Neural Style Transfer - Generative Adversarial Network
44. Generation: Neural Style Transfer Content of an image fused with style of another image *This is illustrative. Not real output from the model(s)
45. Generation: GAN
46. - Training takes a lot of time - More data - Complex model Training: Challenges
47. - Data Parallelism - Model Parallelism Training: Parallelism
48. Training: Data parallelization http://timdettmers.com/2014/10/09/deep-learning-data-parallelism/ Need to synchronize gradients during backward pass MXNet uses data parallelism by default
49. Training: Model parallelization Need to synchronize for both forward pass and backward pass http://timdettmers.com/2014/10/09/deep-learning-data-parallelism/
50. Outline for today... 1. Why deep learning now? 2. How to adopt a practical approach? a. Learning b. Data c. Tools & Deploy 3. Where do you go from here?
51. Code : Tools - Hardware - Software
52. Hardware: GPU (no brainer ) - Single GPU? - Cluster? - Cloud? - Build your own? It depends on the problem(s)
53. Software: Computational Graph Static Model Architecture Defined Computational Graph compiled Model trained Dynamic Model Architecture Defined Computational Graph created for every run
54. Software: Tensorflow Vs PyTorch Tensorflow: good for productionizing PyTorch: good for rapid prototyping of ideas Some pointers on making the choice: - Tensorflow does have eager execution and fold - but PyTorch is more Pythonic and quite popular with researchers - Horovod is quite good for distributed training on tensorflow - MXNet has distributed training at its core - but no widespread adoption yet
55. Deploy: Production - Cloud (Rest API - Tensorflow Serving, Flask) - Edge (CoreML, Tensorflow Lite) - Browser (deeplearn.js / mxnet.js / keras.js)
56. Deploy: Cloud vs Edge vs Browser - Easier to update on cloud - Faster prediction on edge - Energy consumption ! - Model size is HUGE! - Pruning - Quantization (typical: 8 bit) - SqueezeNet
57. Outline for today... 1. Why deep learning now? 2. How to adopt a practical approach? a. Learning b. Data c. Tools & Deploy 3. Where do you go from here?
58. Where do you go from here? - Learn deep learning: resource link - Practice, Practice, Practice! - Take iterative approach
59. Deep Learning A practitioner’s perspective Amit Kapoor amitkaps.com Bargava Subramanian bargava.com