video understanding twitter
play

VIDEO UNDERSTANDING @ TWITTER C O U R T E S Y O F C O R T E X - PowerPoint PPT Presentation

VIDEO UNDERSTANDING @ TWITTER C O U R T E S Y O F C O R T E X USER PROTECTION T W I T T E R C O R T E X T W I T T E R C O R T E X CONTENT UNDERSTANDING T W I T T E R C O R T E X T W I T T E R C O R T E X T W I T T E R C O R T E X


  1. VIDEO UNDERSTANDING @ TWITTER C O U R T E S Y O F C O R T E X

  2. USER PROTECTION T W I T T E R C O R T E X T W I T T E R C O R T E X

  3. CONTENT UNDERSTANDING T W I T T E R C O R T E X T W I T T E R C O R T E X

  4. T W I T T E R C O R T E X

  5. CONTENT INDEXING T W I T T E R C O R T E X

  6. Two dogs sitting on a event poster couch A red car parked in front of a building T W I T T E R C O R T E X

  7. CONTENT RELEVANCE T W I T T E R C O R T E X

  8. T W I T T E R C O R T E X

  9. USER PROTECTION CONTENT UNDERSTANDING CONTENT INDEXING CONTENT RELEVANCE T W I T T E R C O R T E X

  10. VIDEOS — THE COMBINATORIAL CHALLENGE T W I T T E R C O R T E X

  11. T W I T T E R C O R T E X

  12. T W I T T E R C O R T E X

  13. T W I T T E R C O R T E X

  14. Twitter Images Vine Periscope T W I T T E R C O R T E X

  15. FORWARD OR BACKWARD ? T W I T T E R C O R T E X

  16. CLOCKWISE OR COUNTER-CLOCKWISE ? MOTION: FORWARD OR BACKWARD ? T W I T T E R C O R T E X

  17. MULTIMODAL � � � VIDEO � � � AUDIO � � � TEXT T W I T T E R C O R T E X

  18. 0.93933 0.43433 George Toderici and Sudheendra Vijayanarasimhan http://googleresearch.blogspot.com/2015/04/beyond-short-snippets-deep-networks- for.html T W I T T E R C O R T E X

  19. T W I T T E R C O R T E X

  20. RESEARCH ANSWERS T W I T T E R C O R T E X

  21. CNN VARIATIONS VGG SIMONYAN AND ZISSERMAN TWITTER CORTEX T W I T T E R C O R T E X

  22. CNN VARIATIONS Inception Net SZEGEDY AND AL. TWITTER CORTEX T W I T T E R C O R T E X

  23. CNN VARIATIONS Residual Nets KAIMING HE AND AL. TWITTER CORTEX T W I T T E R C O R T E X

  24. RECURRENT NEURAL NETWORKS TWITTER CORTEX T W I T T E R C O R T E X

  25. T W I T T E R C O R T E X

  26. ? T W I T T E R C O R T E X

  27. CORTEX SOLUTIONS T W I T T E R C O R T E X

  28. TAXONOMY EFFORT FAST PROTOTYPING SCALABILITY EASY DEPLOYMENT T W I T T E R C O R T E X

  29. TAXONOMY EFFORT FAST PROTOTYPING SCALABILITY EASY DEPLOYMENT T W I T T E R C O R T E X

  30. TAXONOMY EFFORT PRODUCT � MANAGER Defines a taxonomy LABELING � TRAINING MODEL DEPLOYMENT T W I T T E R C O R T E X

  31. TAXONOMY EFFORT PRODUCT � MANAGER Defines a taxonomy LABELING � TRAINING MODEL DEPLOYMENT T W I T T E R C O R T E X

  32. TAXONOMY EFFORT FAST PROTOTYPING SCALABILITY EASY DEPLOYMENT T W I T T E R C O R T E X

  33. CORTEX PROCESS 1. Data gathering 4. Production Service 
 & Traffic 3. Model Training 2. Labeling T W I T T E R C O R T E X

  34. CORTEX REAL PROCESS D MODEL PROTOTYPING L E PROJECT IDEA A Hey, let’s build a model that B B generates video depending U on tweet contents and user’s E mood! G L G I I N N PREVIOUS SLIDE G MODEL TRAINING G That means you’re lucky T W I T T E R C O R T E X

  35. SOFTWARE NEEDS HARDWARE DATA Low latency storage PROCESSING DATABASE Data download g n i d o c e d o e d i v d DISTRIBUTED e s s e r p TRAINING m o C Training speed NETWORK VIDEO CONVOLUTIONAL CLASSIFICATION LAYERS Computation + Memory hungry CPU Variable temporal sequence FULLY CONNECTED word dictionary LAYERS MULTI-GPU Memory hungry LSTM LOOKUP TABLE T W I T T E R C O R T E X

  36. SOFTWARE NEEDS HARDWARE DATA Low latency storage PROCESSING DATABASE Data download g n i d o c e d o e d i v d DISTRIBUTED e s s e r p TRAINING m o C Training speed NETWORK VIDEO CONVOLUTIONAL CLASSIFICATION LAYERS Computation + Memory hungry CPU Variable temporal sequence FULLY CONNECTED word dictionary LAYERS MULTI-GPU Memory hungry LSTM LOOKUP TABLE T W I T T E R C O R T E X

  37. CORTEX 1.0 T W I T T E R C O R T E X

  38. Target Output Input FORWARD Parameters FORWARD Loss output GradParameters BACKWARD GradOutput ACCGRADPARAMETERS nn.Criterion GradInput BACKWARD nn.Module T W I T T E R C O R T E X

  39. Input GradInput Parameters GradParameters AUTOGRAD FUNCTION Target Loss output Some other torch-autograd function* stuff T W I T T E R C O R T E X

  40. CASCADED DEEP NETS FOR BINARY CLASSIFICATION —— DYNAMIC DAG torch-autograd function T W I T T E R C O R T E X

  41. GITHUB.COM/TWITTER/TORCH-AUTOGRAD T W I T T E R C O R T E X

  42. TAXONOMY EFFORT FAST PROTOTYPING SCALABILITY EASY DEPLOYMENT T W I T T E R C O R T E X

  43. LARGE SCALE DATASET T W I T T E R C O R T E X

  44. SOFTWARE NEEDS HARDWARE DATA Low latency storage PROCESSING DATABASE Data download g n i d o c e d o e d i v d DISTRIBUTED e s s e r p TRAINING m o C Training speed NETWORK VIDEO CONVOLUTIONAL CLASSIFICATION LAYERS Computation + Memory hungry CPU Variable temporal sequence FULLY CONNECTED word dictionary LAYERS MULTI-GPU Memory hungry LSTM LOOKUP TABLE T W I T T E R C O R T E X

  45. DATA STORAGES THREADS MODEL INPUTS HDFS Download/Decode 1 Download/Decode 2 Download/Decode 2 VIDEO BATCH BLOB STORAGE 64*N*3*256*256 … … . . . . ……… Download/Decode 64 Download/Decode 64 T W I T T E R C O R T E X

  46. DATA STORAGES THREADS MODEL INPUTS HDFS Download/Decode 1 Download/Decode 2 Download/Decode 2 VIDEO BATCH BLOB STORAGE 64*N*3*256*256 … . . ……… Download/Decode 64 Download/Decode 64 T W I T T E R C O R T E X

  47. UNLIMITED SAMPLING CAPABILITY T W I T T E R C O R T E X

  48. GITHUB.COM/TWITTER/TORCH-DATASET T W I T T E R C O R T E X

  49. LARGE SCALE TRAINING T W I T T E R C O R T E X

  50. SOFTWARE NEEDS HARDWARE DATA Low latency storage PROCESSING DATABASE Data download g n i d o c e d o e d i v d DISTRIBUTED e s s e r p TRAINING m o C Training speed NETWORK VIDEO CONVOLUTIONAL CLASSIFICATION LAYERS Computation + Memory hungry CPU Variable temporal sequence FULLY CONNECTED word dictionary LAYERS MULTI-GPU Memory hungry LSTM LOOKUP TABLE T W I T T E R C O R T E X

  51. GITHUB.COM/TWITTER/TORCH-DISTLEARN T W I T T E R C O R T E X

  52. LARGE SCALE INFERENCE T W I T T E R C O R T E X

  53. T W I T T E R C O R T E X

  54. PREDICTIONS GPU CLUSTER T W I T T E R C O R T E X

  55. TAXONOMY EFFORT FAST PROTOTYPING SCALABILITY EASY DEPLOYMENT T W I T T E R C O R T E X

  56. ANYBODY CAN TRAIN PLAY DEPLOY T W I T T E R C O R T E X

  57. T W I T T E R C O R T E X

  58. NEW PROJECT PROTOTYPING LARGE SCALE DEPLOYMENT CLUSTER T W I T T E R C O R T E X

  59. NEW PROJECT PROTOTYPING < 1 WEEKS LARGE SCALE DEPLOYMENT CLUSTER T W I T T E R C O R T E X

  60. THANK YOU !! T W I T T E R C O R T E X

  61. SLIDES SKETCH • Twitter Cortex mission statements: • User protection • Content Understanding • Content indexing • Content relevance T W I T T E R C O R T E X

  62. SLIDES SKETCH • Focus on video: how is it challenging? • Pixels distribution (darker, blurry) • Combinatorial challenge (motion, streaming, size, mulitmodal) T W I T T E R C O R T E X

  63. SLIDES SKETCH • Focus on video: what research tells us • CNNs • RNNs • Multimodal? T W I T T E R C O R T E X

  64. SLIDES SKETCH • Focus on video: Cortex industrial answer • Smart taxonomy [organization] (fast labeling, feedback) • Fast Prototyping [software] (torch-autograd) • Scalability [software + hardware] (torch-dataset, torch-distlearn) • Fast deployment [software] (embedded torch, no extra step) T W I T T E R C O R T E X

Recommend


More recommend