tensorflow-v2

This repo is a Tensorflow course starting from scratch. The inspiration coming from awesome guys to whom I say thanks a lot.

• AntonMu
• Aymeric Damien
• YunYang1994
• zzh8829
• Anton Mu

Prerequisites

Python environment is required and correctly configured in order o use these exercises.

Please check:

• Anaconda.
• Nvidia Cuda 10.1
• Google TensorFlow 2.0
• Keras

Guides

tflearn

Example and Tutorials

• Git Tensorflow

• Aymeric Damien
• YunYang1994 Examples

• Tensorflow Yolo v3

• zzh8829
• YunYang1994 yolov3

Contents

1 - Introduction

• 01-helloworld.py: It is a simple check to evaluate if Tensorflow is correctly installed.
• 02-helloworld-gpu.py: Same as the item above but checking the GPU support (Nvidia Cuda GPU is required).
• 03-constants.py: Basic example about Tensorflow tensors. Simple Linear Algebra ops are executed as well.

2 - Keras - The models

• 04-linear-model.py: The basic linear model.
• 05-tensorboard.py: How write trace to Tensorboard.
• 06-load-traininset.py: Load the trainingset data from csv file.
• 07-save-weights.py: How to save the training weights and load it when available.
• 08-use-model-class.py: Complete example to training, save the weights and load it if present.

3 - Standard Datasets

• 09-mnist-dataset.py: The helloworld MINIST dataset.
• 10-cifar-dataset.py: The CIFAR-10 dataset.

4 - Images Datasets

• 11-images-dataset.py: Run CNN on images classified by folder name.
• 12-images-dataset-cache.py: Same as the above test!!!
• 13-images-dataset-local.py: Same as the above but from local folder
• 14-convert-to-tensorflow-lite.py: Convert the model to tensorflow lite model available to be used on mobile device

How to install from scratch

Nvidia GPU support CUDA Toolkit 11.6 Update 1 Downloads cuda_11.6.0_511.23_windows Download cuDNN v8.3.2 (January 26th, 2021), for CUDA 11.0,11.1 and 11.2 cudnn-windows-x86_64-8.3.2.44_cuda11.5-archive

python-3.9.7.exe (DO not use python-3.8.5.exe)

• Check Nvidia cuda version

nvcc --version

Checked version: nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2021 NVIDIA Corporation Built on Fri_Dec_17_18:28:54_Pacific_Standard_Time_2021 Cuda compilation tools, release 11.6, V11.6.55 Build cuda_11.6.r11.6/compiler.30794723_0

Preliminary checks

python --version
conda --version
pip3 --version

# Checked version:
Python 3.9 (3.9.7)
conda 4.10.3
pip 22.0.3

• Create a log folder like c:\log

• Create the conda environment

conda create --name tensorflow python=3.9
conda activate tensorflow
conda install git
python -m pip install --upgrade pip
pip install -r requirements.txt

• Run tensorboard

(tensorflow) tensorboard --logdir=c:\log --port 6006

Open a web browser http://localhost:6006

Tips and Tricks

Run following commands before run jupyter at https://github.com/tensorflow/models

https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md

conda install git
pip3 install "git+https://github.com/philferriere/cocoapi.git#egg=pycocotools&subdirectory=PythonAPI"

How to use AWS EC2 intances

Connect to AWS Select the environment Update the system and clone the repo

sudo apt-get update
git clone https://github.com/AntonMu/TrainYourOwnYOLO.git