Tensorflow

install[]

git checkout r1.9 of tensorflow and install gcc 4.8.5 to compile tensorflow from source for cuda9.2. Remove all numpy and cython packages, allow bazel to install it. https://github.com/numpy/numpy/issues/11871 and https://github.com/cmusatyalab/openface/issues/30 for numpy failing to load arrays when compiling numpy and cython from source. See Opencv_install_script#gcc_version, same issue with opencv, use gcc 4.8.5 for cuda9.2 , cuda 10 use gcc 7

https://www.python36.com/how-to-install-tensorflow-gpu-with-cuda-9-2-for-python-on-ubuntu/2/
https://gist.github.com/Brainiarc7/6d6c3f23ea057775b72c52817759b25c
https://devtalk.nvidia.com/default/topic/845363/libcublas-so-7-0-cannot-open-shared-object-file/
https://stackoverflow.com/questions/13428910/how-to-set-the-environmental-variable-ld-library-path-in-linux.
https://pastebin.com/jSCYQ5kg cython ModuleNode.py hack to get tensorflow installed with cuda9.2. the numpy , cython build from source cannot be used by tensorflow. remove all packages

Caffe to tensorflow[]

convert Caffe berkeley vision models to tensorflow, which allows easier python based neural networks. https://github.com/ethereon/caffe-tensorflow , http://www.cs.toronto.edu/~guerzhoy/tf_alexnet/ linked from David Silver http://medium.com Three front ends available for tensorflow

https://github.com/tensorflow/tensorflow/tree/master/tensorflow/contrib/slim tensorflow slim
https://keras.io/
http://tflearn.org/ David Silver's preferred solution

resources[]

github khanhnamle1994 course solutions for university and Udacity

https://github.com/fchollet/deep-learning-with-python-notebooks This repository contains Jupyter notebooks implementing the code samples found in the book Deep Learning with Python (Manning Publications). Note that the original text of the book features far more content than you will find in these notebooks, in particular further explanations and figures. Here we have only included the code samples themselves and immediately related surrounding comments.

Two minute papers
https://github.com/pjreddie/TopDeepLearning from pd reddie author of Yolo

https://github.com/trekhleb/homemade-machine-learning/blob/master/README.md his repository contains examples of popular machine learning algorithms implemented in Python with mathematics behind them being explained. Each algorithm has interactive Jupyter Notebook demo that allows you to play with training data, algorithms configurations and immediately see the results, charts and predictions right in your browser. In most cases the explanations are based on this great machine learning course by Andrew Ng.

https://wiki.tum.de/display/lfdv/Tensorflow#Tensorflow-Variables

https://pprc.qmul.ac.uk/~bevan/statistics/TensorFlow_tutorial.pdf

https://github.com/daltskin/MachineLearningFromModelToProduction

https://github.com/Hvass-Labs/TensorFlow-Tutorials

https://github.com/nfmcclure/tensorflow_cookbook code from the book tensorflow cookbook.

jtoy[]

https://github.com/jtoy/awesome-tensorflow A curated list of TensorFlow experiments, libraries, and projects. https://github.com/pkmital/tensorflow_tutorials links to https://www.kadenze.com/courses/creative-applications-of-deep-learning-with-tensorflow-iv/info

tensor[]

Neural papers with code

https://github.com/astorfi/TensorFlow-World#why-use-tensorflow

https://github.com/aymericdamien/TensorFlow-Examples 10000 forks github tutorial linked from astorfi

https://unsupervisedmethods.com/my-curated-list-of-ai-and-machine-learning-resources-from-around-the-web-9a97823b8524

Yolo[]

Yolo

https://github.com/xslittlegrass/CarND-Vehicle-Detection Detecting vehicles in a video stream is an object detection problem. An object detection problem can be approached as either a classification problem or a regression problem. As a classification problem, the image are divided into small patches, each of which will be run through a classifier to determine whether there are objects in the patch. Then the bounding boxes will be assigned to locate around patches that are classified with high probability of present of an object. In the regression approach, the whole image will be run through a convolutional neural network to directly generate one or more bounding boxes for objects in the images.

https://github.com/allanzelener/YAD2K You only look once, but you reimplement neural nets over and over again. YAD2K is a 90% Keras/10% Tensorflow implementation of YOLO_v2. Original paper: YOLO9000: Better, Faster, Stronger by Joseph Redmond and Ali Farhadi. https://arxiv.org/abs/1612.08242

https://github.com/sunshineatnoon/Darknet.keras/ , http://www.robots.ox.ac.uk/~joao/ , https://github.com/vojirt/kcf Kernelized correlation filters. http://www.robots.ox.ac.uk/~joao/circulant/ Both KCF and DCF outperform top-ranking trackers such as Struck or TLD on a 50 videos benchmark, despite running at hundreds of frames-per-second, and being implemented in a few lines of code. To encourage further developments, our tracking framework was made open-source. http://rodrigob.github.io/#code is the python port, but C++ is also available. https://github.com/rodrigob/barinova_pedestrians_detection This a linux port of the original code provided by Olga Barinova from the Vision Group at Moscow State University, 2010. Please visit the project website for more details. This derivative work follows the Microsoft Research Shared Source license, which allows only non-comercial usage.(meaning commercial companies will have to pay millions to use it in a commercial product) . The FSF Stallman newspeak(GPL and BSD) on GPL doesn't make it clear that the copyright holders can arbitrarily wave this non commercial restriction if you pay them lots of money. Pedestrians detection using Hough forests is a derivative work from http://graphics.cs.msu.ru/en/science/research/machinelearning/hough. o detect multiple objects of interest, the methods based on Hough transform use non-maxima supression or mode seeking in order to locate and to distinguish peaks in Hough images. Such postprocessing requires tuning of extra parameters and is often fragile, especially when objects of interest tend to be closely located. In the paper, we develop a new probabilistic framework that is in many ways related to Hough transform, sharing its simplicity and wide applicability. At the same time, the framework bypasses the problem of multiple peaks identification in Hough images, and permits detection of multiple objects without invoking nonmaximum suppression heuristics. As a result, the experiments demonstrate a significant improvement in detection accuracy both for the classical task of straight line detection and for a more modern category-level (pedestrian) detection problem.

SSD[]

https://www.youtube.com/watch?v=6q-DBCPROA8 single shot multibox detector. Provides comparable accuracy to explicit region proposal methods (such as Faster R-CNN) but is much faster and thus better suited for real-time applications.

https://github.com/weiliu89/caffe/tree/ssd

paper[]

http://cs229.stanford.edu/proj2016/report/BuhlerLambertVilim-CS229FinalProjectReport.pdf We reimplement YOLO, a fast, accurate object detector, in TensorFlow. To perform inference, we leverage weights that were trained for over one week on GPUs using Ima- geNet data, a publicly-available dataset containing several million natural images. We demonstrate the ability to repro- duce detections comparable with the original implementa- tion. We learn the parameters of the network and compare mean average precision computed from pre-trained network parameters. Furthermore, we propose a post-processing scheme to perform real-time object tracking in live video.

Redmon et al. ’s work is especially notable for two major strengths. First, their model solves in an end-to-end fashion what was considered in the not-far-distant past two separate problems in computer vision literature: object detection and object classification. Second, their model presents an efficient solution to an enduring problem in computer vision: how does one go about producing an arbitrary number of detections in an image while using fixed dimensional input, output, and labels? YOLO avoids computationally expensive region proposal steps that detectors like Fast R-CNN[4] and Faster-RCNN[14] require. However, since the time of YOLO’s publication, newer models such as Single-Shot Multi-Box Detectors [9] seem to offer improvement in mAP with reduced GPU inference time [6]. YOLO uses grid cells as anchors to detections, much like Faster R-CNN and Multi-Box.

tutorials[]

https://www.youtube.com/watch?v=cKxRvEZd3Mw&t=2s Google developer channel.

https://github.com/Newmu/Theano-Tutorials theano tuorial. see https://www.youtube.com/watch?v=MDP9FfsNx60

https://github.com/jostmey/NakedTensor basic tensorflow tutorial

https://github.com/llSourcell?tab=repositories Siraj Raval

Martin Gorner[]

https://codelabs.developers.google.com/codelabs/cloud-tensorflow-mnist/#0, https://www.youtube.com/watch?v=vq2nnJ4g6N0#t=2.971187

https://goo.gl/UuN41S slides with video
https://goo.gl/pHeXe7 slides with youtube video.

links[]

http://rimstar.org/science_electronics_projects/backpropagation_neural_network_software_3_layer.htm

Pytorch
Nervanasys Video activity learning, use neural net to detect a carjacking for example or any other activity.
Keras
Neural papers with code