On-device machine learning: TensorFlow on Android (Google Cloud Next ’17) November 24, 2019 35 By Stanley Isaacs CategoryArticles BlogTags#GoogleNext17 Cloud cloud next custome dataset data dataset deep learning deep learning model GCP Google cloud local prediction machine learning model Mobile mobile devices TensorFlow 35 Comments swarathesh addanki says: March 15, 2017 at 10:33 pm can you post the code? Reply Yu-Ting Cheng says: March 17, 2017 at 3:43 am Your explanation of CNN is the best and simplest I have ever heard. Thank you. Reply swarathesh addanki says: March 21, 2017 at 6:13 pm can you make detailed tutorial on how to do ? Reply Brian Chu says: March 24, 2017 at 5:12 pm awesome talk at 1.25x speed Reply citizen000 says: April 8, 2017 at 8:48 pm Very inspiring talk. Great work ! Reply A Smith says: April 26, 2017 at 10:26 pm Reese's, damn it, not "Reesies". Jesus. Reply Hans Baier says: April 26, 2017 at 11:46 pm Where is the code lab you talked about? I did not understand how to downsize the model. Why won't we need the feature extraction layers? Reply JamesUS100 says: May 1, 2017 at 2:14 am thanks for the clarity on a complex topic Reply Rohit N says: May 1, 2017 at 4:37 pm Google Cloud Next'17 .. has PPT issues.. Reply Vetri Selvan says: May 8, 2017 at 5:15 pm great topic Reply Abhijeet Bhanjadeo says: May 20, 2017 at 3:12 am Valuable content with super easy explanation for CNN.Here are some key take away from this video:What is Convolutional Neural Network(CNN) https://youtu.be/EnFyneRScQ8?t=8m2sHow to Build & Train the Model with ease(images from videos) https://youtu.be/EnFyneRScQ8?t=5m25sPipeline for the Classification App https://youtu.be/EnFyneRScQ8?t=7m26sDownsizing the Model(Graph Transform Tool) for Mobile Device https://youtu.be/EnFyneRScQ8?t=12m57sHow to put the model inside Mobile Application(Architecture) https://youtu.be/EnFyneRScQ8?t=14m41sReducing the problem size by consider the usage(same thing on Raspberry Pi) https://youtu.be/EnFyneRScQ8?t=21m38sFinally Resources https://youtu.be/EnFyneRScQ8?t=23m39s Reply station says: June 6, 2017 at 4:55 am too bad they didn't serve hotdogs for snacks Reply Will Tesler says: June 27, 2017 at 10:03 pm Awesome presentation. Reply Mostafa Elzoghbi says: June 29, 2017 at 8:03 pm love it! Reply Alvar Lagerlöf says: July 1, 2017 at 7:26 pm I used that code lab. It's awesome! Now I can recognize a 100 different dog species with 70-100% accuracy! Reply Code Developer says: July 29, 2017 at 3:05 pm You explained simply and easy to understand about machine learning. This is a awesome to identify the things. Is it possible that how we can implement it in android? Help would be very thankful. Thanks Reply Ritesh Kumar Maurya says: August 31, 2017 at 10:57 am How to strip a video into pictures for training on that data? Reply Karthik Arumugham says: September 7, 2017 at 4:45 pm @Yufeng Guo Great talk. Loved the use of jigsaw puzzle for CNN explanation. Thank you! Reply ILikeWeatherGuy says: September 13, 2017 at 4:37 pm uh cant wait until computer processors are fast enough to locally do a bunch of these layers Reply thoughtsmith innovation says: October 9, 2017 at 10:05 am The codelab that you mentioned is using mobilenet not inception-v3. should i just change the name when i am referring the architecture? or is there some different value? Reply Sangho Bose says: October 11, 2017 at 4:50 am Video -> images : Awesome Reply Mark Silla says: November 6, 2017 at 4:28 am Was he working out before this talk? A lot of panting. Great talk though. It was simple yet informative. Reply moonSpawn says: December 8, 2017 at 5:42 pm 6:33 that got existential pretty fast, watching YouTube on YouTube. Reply Dr Tune says: December 15, 2017 at 7:16 pm Great talk, thanks Reply Nikola Marić says: January 5, 2018 at 7:09 pm In Japan, even farmers can make robots… O.o Reply 毛笔小鑫 says: January 19, 2018 at 11:59 am whos guoyufeng Reply Shravil Potdar says: February 3, 2018 at 8:19 am Amazing 🙂 Reply Firefox Metzger says: February 4, 2018 at 5:45 pm I love the jigsaw puzzle analogy Reply Qichuan Zhang says: February 21, 2018 at 3:15 am One of the best ML talks I have ever had. Looking forward to seeing more from Yufeng. Reply Jieru Zhang says: April 17, 2018 at 9:52 am Brilliant! Split a short video into many images, and it is a method of data enhancement I would say. Reply Jack Liu says: April 28, 2018 at 2:41 am I really have learned a great lesson from it. Thanks! Reply teja polisetty says: February 6, 2019 at 11:40 pm But you did't tell how to deploy ML model into android Reply Systems Planet says: April 18, 2019 at 12:22 pm – During training, how does TensorFlow distinguish the object from its background? – Do you have to manually mask out the background colors? – Would that improve recognition? – If I trained it it to detect 60 colored shapes,could it detect 3 red of the same type at the same time? – Is it possible to start without using google cloud? Can I train at home. – How would you train it to recognize individual squares in a Rubik's Cube? One square at a time? – Is it possible to retrieve the small clipped image data that was recognized? Eg individual squares? Reply Sean Spicer says: April 24, 2019 at 10:23 pm who took the thumbnail picture? google, the <pronoun> who created this thumbnail has either 1) received a promotion and a raise by u 2) left due to being unrecognized. Reply Dawood Muzammil says: July 18, 2019 at 8:38 am There are types of cucumbers?!?!?!?!?! Reply Leave a Reply Cancel reply Your email address will not be published. Required fields are marked *Comment Name * Email * Save my name, email, and website in this browser for the next time I comment.