Using Convolutional Neural Networks and TensorFlow for Image Classification
At the recent NYC TensorFlow meetup, it was discussed how to train an image classifier with TensorFlow. From this blog post, you will learn about the challenges of image classification / search and how to address them with convolutional neural networks.
Training an image classifier with TensorFlow
In his session, Scott Thompson of KONTOR talked about:
- image classification
- visual search across the Gilt catalog
- and how convolutional neural networks can enhance the process
When setting up image features (edge detection, color histogram, pixel segmentation, etc.) for classification or search, the job is mostly manual and takes time. According to Scott, convolutional neural networks provide a black box to construct image features. He also outlined some advantages of using a pre-trained model:
- it’s faster (as it’s pre-trained already)
- it’s cheaper (no need for GPU farm)
- it generalizes (avoids overfitting)
You can also look through the full slides:
Join the meetup group to stay informed about the upcoming events.
About the speaker
Scott Thompson is a full-stack engineer who enjoys focusing on large scale data mining, predictions, and recommendations. He is particularly interested in Scala, Clojure, and ClojureScript applications.
Performance Benchmark: Caffe, Deeplearning4j, TensorFlow, Theano, and Torch
To stay tuned with the latest updates, subscribe to our blog or follow @altoros.