Take Apart and Rebuild a Neural Network for OCR with TensorFlow
summary
A practical presentation on building and rebuilding a neural network for optical character recognition (OCR) using TensorFlow. It covers lessons learned from designing, training, and iterating on a TensorFlow-based OCR model, including practical challenges and improvements encountered during development.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 5:06Client problem definition
- 7:58Existing solutions analysis
- 10:25Black box approach choice
- 13:59Initial failures and lessons
- 15:44Problem simplification
- 17:30Modular network architecture
- 19:39Convolutional layers explained
- 24:48Concept testing and validation
- 27:57Bidirectional recurrent networks
- 31:31Visual attention mechanism
- 34:20Results and metrics
- 37:13Infrastructure and deployment
- 38:16Lessons learned and best practices
- 41:48Experience feedback and conclusion
related talks
53mNowoczesne uczenie sieci neurownowych w TensorFlow
01h 06mDeep Learning with TensorFlow
01h 29mComputer Vision: An (Un?)Expected Journey, with Keras and Tensorflow
40mTensorflow: Designing neural networks
43mTensorFlow and Deep Learning without a PhD
44mHello Deep Learning
40mDeep Learning and Optical Character Recognition
01h 50mNeural Networks: Walkthrough