My first steps on Deep Learning with Keras
summary
15-minute demo shows Python code using Keras with VGG16 and transfer learning: freeze the base, retrain only last layer on ~2% of catalog images with 40% held for validation, yielding >95% accuracy on clothing classes. Image search indexes precomputed feature vectors and returns nearest neighbors by cosine distance on feature vectors. Notes practical failures: mannequin/background bias and need for preprocessing, category filtering and more data to improve real-world results.
AI-written summary · about
chapters (4)
related talks
35mDeconstructing Deep Learning
01h 01mA practical guide to deep learning
42mDeep Learning for Computer Vision
01h 39mGoogle Colaboratory and Python: Deep Learning basics for CV tasks
01h 29mComputer Vision: An (Un?)Expected Journey, with Keras and Tensorflow
32mYou Can Do Deep Learning!
47mDeep Learning in Computer Vision
01h 55mGoogle Colaboratory and Python: Deep Learning basics for CV tasks