This repo contains the code for:
- a credit card image generator
- an android app developed for my bachelor thesis
- the latex files to generate the final thesis report
Speed comparison of convolutional neural networks in TensorFlow
TensorFlow is an open-source library developed by Google for machine learning applications. It can be run in different environments such as desktop computers, servers or even browsers. For running TensorFlow on mobile devices there exist TensorFlow Mobile and the recently launched TensorFlow Lite. The latter is an improved version of the former, which claims to have several performance advan- tages over its predecessor and is meant to substitute it on the long run. The goal of this study is to test and compare the performance of this two libraries when run- ning a convolutional neural network trained to analyze pictures of credit cards. An application was programmed to run the same network with both versions of TensorFlow and installed on several devices to measure their performance. The collected empirical data showed that TensorFlow Lite’s performance is still lower than its counterpart’s when running convolutional neural networks. This is due to the early stage of development of the library, which is not optimized enough yet.
The final paper can be read here