README

NEWS

Date	News	Version
Sept 2019	face recognition (insight face) was released for inferencing (STABLE), for training will available in the future version	>= v0.7.4
August 2019	deeplab semantic segmentation (PREVIEW version)was released	>= v0.6.10
Mei 2019	yolov3 object detection (STABLE version) was relesed	> v0.5.1
April 2019	Unet-segmentation (PREVIEW version)	v0.4.18

Tensorflow Compatibility

Tensorflow version	Simple-Tensor Version
1.4.1 - 1.1	<=v0.4.0
1.13.1 and 1.15.0	>=v0.4.3

ABOUT PROJECT

This project is a simplification of tensorflow operations and related projects

DEPENDENCIES

1. Tensorflow (1.4.1 - 1.15.0) pip3 install tensorflow-gpu
2. Comdutils pip3 install comdutils
3. opencv-python pip3 install opencv-python
4. numpy

HOW TO USE

Installing The Package

pip3installsimple-tensor

Import The Package

Tensor Operations

importtensorflowastf# tensor operationsfromsimple_tensor.tensor_operationsimport*# tensor lossesfromsimple_tensor.tensor_lossesimport*# tensor metricsfromsimple_tensor.tensor_metricsimport*

This packages contains tensor operations (conv2d, conv1d, depthwise conv2d, fully connected, conv2d transpose), tensor losses (softmax & sigmoid cross entropy, MSE), and tensor metrics (accuracy). For more detail documentations about tensor operations, visit this page

Transfer Learning Package (Image Classification)

importtensorflowastffromsimple_tensor.transfer_learning.inception_utilsimport*fromsimple_tensor.transfer_learning.inception_v4import*

This package contains a library of tensorflow implementation of Inception-v4 for image classification. Densenet, Resnet, and VGG will be added in the future version. For more detail documentations about transfer learning package, visit this page

(img source: link)

Object Detector Package

importtensorflowastffromsimple_tensor.object_detector.detector_utilsimport*fromsimple_tensor.object_detector.yoloimportYolo

This package contains a library of tensorflow implementation of Yolov3 (training and inferencing). You can customize your yolo detector with four types of network ("big", 'medium", "small", "very_small"). For more detail documentations about object detector package (yolov3), visit this page.

(img source: pjreddie)

Unet Segmentation Package

importtensorflowastffromsimple_tensor.segmentation.unetimportUNet

This package contains the tensorflow implementation of U-net for semantic segmentation. For more detail, visit this page

(img source: internal)

Face Recognition Package (Insightface)

importtensorflowastffromsimple_tensor.face_recog.insight_faceimport*

This package contains the tensorflow implementation of insight face. This repo is refractoring this github link. For more detail documentation, visit this page

LSTM Package

stillonprogress ....

Convert Keras Model to Tensorflow Serving

importtensorflowastffromsimple_tensor.convertimport*

DOCKER

We already prepared the all in one docker for computer vision and deep learning libraries, including tensorflow 1.12, Opencv3.4.2 and contrib, CUDA 9, CUDNN 7, Keras, jupyter, numpy, sklearn, scipy, statsmodel, pandas, matplotlib, seaborn, flask, gunicorn etc. See the list of dockerfile below:

Docker: Ubuntu 16.04 with GPU (Cuda 9, cudnn 7.2) [TESTED]

• https://github.com/fatchur/Opencv-contribt-Tensorflow-GPU-DS-Tools-Docker/tree/master/docker-16.04

Docker: Ubuntu 18.04 with GPU (Cuda 9, cudnn 7.2)

• https://github.com/fatchur/Opencv-contribt-Tensorflow-GPU-DS-Tools-Docker/tree/master/docker-18.04

Docker: Ubuntu 16.04 without GPU (Cuda 9, cudnn 7.2) [TESTED]

• https://github.com/fatchur/Opencv-contribt-Tensorflow-GPU-DS-Tools-Docker/tree/without_gpu/docker-16.04

Docker: Ubuntu 18.04 without GPU (Cuda 9, cudnn 7.2) [TESTED]

• https://github.com/fatchur/Opencv-contribt-Tensorflow-GPU-DS-Tools-Docker/tree/without_gpu/docker-18.04