2. Great time to be in Deep Learning + Computer Vision!
● Lots of publicly available data/code
○ ImageNet, Pascal VOC, Github
● Very good performance
○ ImageNet, LFW
● Several good deep learning toolkits available
○ Caffe, Torch, TensorFlow
● Cloud resources
○ AWS, Google Cloud
7. Fanreel
● UGC (user-generated content) and editorial
imagery are more engaging and drive more
revenue than stock product photography
● But tagging products is slow, tedious, and
error-prone
8. ● Automatically tag products
○ Cut down on workload
● Other applications
○ Show related products
○ Find cheaper versions of high-end products
○ Find complementary products
Intelligent Product Tagging [blog post]
12. Intelligent Product Tagging
● Detection: Find the products in a query image
● Search: Find the best match for each product
● State-of-the-art uses deep learning for both
components
22. Intelligent Product Tagging
● Detection: Find the products in a query image
● Search: Find the best match for each product
● State-of-the-art uses deep learning for both
components
39. Generative Adversarial Networks [NIPS 2014]
“There are many interesting recent development in
deep learning…The most important one, in my
opinion, is adversarial training (also called GAN for
Generative Adversarial Networks). This, and the
variations that are now being proposed is the most
interesting idea in the last 10 years in ML, in my
opinion.” – Yann LeCun [link]
40. Generative Adversarial Networks [NIPS 2014]
● Most CNN models are discriminative
● GAN framework
○ Generative model
○ Discriminative model
● Analogous to counterfeiters vs police