KeyMe Releases Two Machine Learning Models
Hi everyone, Kris Borer here, CTO of KeyMe. I am very excited to announce the latest open source software contributions from the KeyMe engineering team. We have released source code for two machine learning models that can be found on our Github repository.
The first is our implementation of RefineNet in TensorFlow, based on the paper by Guosheng Lin, et al. RefineNet is a semantic segmentation model that can use fine-grained features when extracting parts of an image. This gives us the extreme precision that we need for key duplication, and we hope it will be useful for your deep learning applications as well.
The second model we have released is our implementation of deep embedded clustering in TensorFlow. DEC is a particularly effective method of grouping data points because it learns an optimized representation of the data without supervision. As a bonus, our implementation includes both the original DEC by Junyuan Xie, et al. and Improved DEC by Xifeng Guo, et al. We use deep embedded clustering for identifying rare key types around the world, and hope you find it as helpful as we do.
Machine Learning Engineers
Allow me also to introduce you to the two developers who actually wrote the code. First is Nick Marton who came to us from Rensselaer Polytechnic Institute. Nick is pictured below with Buttercup, one of our machine learning rigs. Nick built Buttercup by hand using a pair of Titan V GPUs and a secret ingredient (love).
Next is Jiayu Wu who came to us from Stanford. Here he is with Creampuff, one of our training machines that sports four GeForce GTX GPUs . We’ve been very happy with both the GTXs and Titan Vs, but if you have other hardware recommendations please let us know.
At KeyMe, we love machine learning and open-source software. We hope this source code can help other practitioners as well as those who want to learn more about deep learning. And, of course, we look forward to releasing more in the future!