Tensorflow Custom Object Detection Course
Step 1 : Tensorflow Installation
Learner have to go with any one of the following Tensorflow CPU or GPU installation.
CPU means: You only have processor in your system. Eg. Intel processor.
GPU means: You only have NVIDIA graphic card along with system’s processor.
(To check you have NVIDIA GPU or not, Right click on your desktop and check if you have
NVIDIA control panel as below)
If you don’t have above option “NVIDIA Control Panel” upon right clicking on desktop,
you only have to follow (a) Tensorflow CPU Installation whereas if you have above
option, go with (b) Tensorflow GPU installation
Note: click on download button wherever its mention for respective installation.
www.extrinsictechnology.com
a. Tensorflow CPU installation
Download Anaconda
conda create -n tfcpu
conda activate tf
conda install python=3.7.0
pip install Cython contextlib2 pillow lxml jupyter matplotlib pandas
opencv-python numpy==1.16.1 flask werkzeug==0.16.0
pip install tensorflow==1.13.1
Check tensorflow is installed properly or not.
Python (3.7.0) (no error should populate after installation)
>>import tensorflow as tf
>>
---------------------------------------------OR-------------------------------------------------
b. Tensorflow GPU Installation
1. Download visual studio 2017 from below link and install only with .NET
installation which will help us to install compatible C++ library.
VS1027 Pro Download
Note: Our course won’t have any connection related to Visual studio. We
are just installing it for MS C++ libraries.
2. Download CUDA toolkit 10 or 9
What is CUDA toolkit?
CUDA® is a parallel computing platform and programming model developed by NVIDIA
for general computing on graphical processing units (GPUs). With CUDA, developers are
able to dramatically speed up computing applications by harnessing the power of GPUs.
www.extrinsictechnology.com
Check CUDA enabled GPUs
https://en.wikipedia.org/wiki/CUDA
Install any one from below depending on your graphic card
a. CUDA 9 with CUDNN 7.3.1
CUDA 9 CUDNN 7.3.1
Install CUDA 9 and copy CUDNN files in NVIDIA GPU computing folder
BIN
Include
X64
----------------------------------------OR------------------------------------------------------
b. CUDA 10 with CUDNN
CUDA 10 CUDNN 7.4.2
Install CUDA 10 and copy CUDNN files in NVIDIA GPU computing folder
BIN, Include, X64
3. Install Miniconda/ANACONDA
Download Anaconda
conda create -n tfgpu
conda activate tfgpu
conda install python=3.7.0
pip install Cython contextlib2 pillow lxml jupyter matplotlib pandas
opencv-python numpy==1.16.1 flask werkzeug==0.16.0
pip install tensorflow-gpu==1.13.1
(Restart the system)
www.extrinsictechnology.com
Check tensorflow is installed properly or not.
Python (3.7.0) (no error should populate after installation)
>>import tensorflow as tf
All in One Download Link
>>
here
Step 2: Tensorflow Framework Installation
Download Tensorflow
Framework
Download Protos
a. Create tensorflow folder at local disk directory
b. Copy and paste folder “models” and “protos” which is downloaded from
above buttons
c. Activate your conda env
Conda activate tfgpu
Navigate to the CD C:/tensorflow/Models/Research>>
(Type below commands one by one and hit ENTER)
"C:\tensorflow\protoc\bin\protoc.exe" object_detection/protos/*.proto --python_out=.
python setup.py install
Set PYTHONPATH=C:\tensorflow\models\research\slim;C:\tensorflow\models\research
www.extrinsictechnology.com
Step 3: Dataset Preparation and Annotation
1. Prepare dataset – before feeding into machine learning model
2. Dataset : combination of different form of same data
Download some free image dataset you can try for learning purpose.
Rice Disease Dataset Fruits Classification
Malaria Dataset Flower Dataset
Yoga Pose Dataset Simpsons Cartoon DS
I am using animal classification
https://www.kaggle.com/ashishsaxena2209/animal-image-datasetdog-cat-and-panda
------------------------------------------------------------------------------------------------------------------------
Image annotation toolkit installation
Download LABEL_IMG
Use below commands to install labelling/annotation tool
pip install PyQt5
pyrcc5 -o libs/resources.py resources.qrc
python labelImg.py
1. Install Image labelling tool
2. Prepare the image dataset
3. Decide classifier and perform annotation
My labelled dataset Dog and Panda
Labelled dataset
DOG and PANDA
www.extrinsictechnology.com
Step 4: Convert labelled dataset into TF-record
(File generated before model training)
1. Download sublime text Download
2. Download python programs for conversions Download
Now its turn to distribute your labelled dataset into two folder
Folder “train” should consist of 90% of entire dataset
Folder “test” should consist of 10% of entire dataset
Use commands
set PYTHONPATH=C:\tensorflow\models\research\slim;C:\tensorflow\models\research
1. XML to CSV
Python xml_to_csv.py
Use above command twice by changing folder name will result in below two files.
>> test.csv
>>train.csv
Paste above csv files in respective folder
2. CSV to TF-RECORD
Move test and train folder to the tensorflow/models/research/object_detection
Also move generate_ftrecord.py file to the same folder
Mention your classifier name (line 39,41)
“dog” “panda”
Navigate tensorflow/models/research/object_detection
Generate test.record and train.record
www.extrinsictechnology.com
set PYTHONPATH=C:\tensorflow\models\research\slim;C:\tensorflow\models\research
Test.record
python generate_tfrecord.py --csv_input=C:/tensorflow/models/research/object_detection/test/test_labels.csv --
output_path=C:/tensorflow/models/research/object_detection/test/test.record
Train.record
python generate_tfrecord.py --csv_input=C:/tensorflow/models/research/object_detection/train/train_labels.csv --
output_path=C:/tensorflow/models/research/object_detection/train/train.record
www.extrinsictechnology.com
Step 5: Train the model
1. Download SSD model file Download
2. Download Labelmap.pbtxt
Download
3. Download SSD Config File Download
Navigate to the ./object_detection/legacy folder
Create “data” folder ------- paste train.record and test.record in it
Create training folder
Edit labelmap.pbtxt with your classifier name (eg. Dog,cat etc) and copy-paste
labelmap.pbtxt to “data” and “training” folder
Paste ssd config file in training folder
EDIT SSD config file with batch size, training step, etc
Copy downloaded SSD model file (120MB) and paste into legacy
www.extrinsictechnology.com
set PYTHONPATH=C:\tensorflow\models\research\slim;C:\tensorflow\models\research
Run below commands to start training (I will suggest to copy first in notepad and then paste to
anaconda prompt)
python train.py --logtostderr --train_dir=training/ --
pipeline_config_path=training/ssd_mobilenet_v1_coco.config
Step 6: GENERATE MODEL GRAPH AND DEPLOYMENT
Copy export_inference_graph.py from object detection folder and paste into legacy dir.
Use below command by editing “model name” and “your_number”
python export_inference_graph.py --input_type image_tensor --pipeline_config_path
training/ssd_mobilenet_v1_coco.config --trained_checkpoint_prefix training/model.ckpt-your_number --
output_directory model_name
Check how model is trained with the help of tensorboard (run below command in legacy folder)
tensorboard --logdir ./ --host=127.0.0.1
if tensorboard error occurred.
pip install tb-nightly
Deployment
Webserver Code Download
Commands
Set flask_env=development
Flask run
Webcam Code Download
Video Feed Code Download
Android Code Download
Entire project – All required files download link
All in One Download Link
here
www.extrinsictechnology.com
Thank you ML enthusiastic for learning this course.
Good Luck for your future!
For more help you can reach out
Instagram: https://www.instagram.com/tf.aakash
Email:
[email protected]Linkedin: https://www.linkedin.com/in/aakashjadhav16197/
YouTube : https://www.youtube.com/c/extrinsictechnology
Whatsapp:
https://chat.whatsapp.com/BUlWF9NLgMY3JsSAMr1ZMt
Website: www.extrinsictechnology.com
Facebook: https://m.facebook.com/extrinsicTech
www.extrinsictechnology.com
