Transcripts
1. Yoga Pose Detection Project introduction: Come ever on to the complete missile learning project course. This course is about yoga post
prediction using resonet. So this project consists
of different modules. We are going to discuss about the concepts like how to import a dataset for our project and how to train our
dataset using Resonet. At last, we are going to
predict the output for our project yoga post
prediction using Resnet. So this project course is very, very useful for the
students who are trying to create resonatPject
for the first time. So let's get started.
2. Yoga Class 1 : Import Packages: Welcome everyone to
the first class. In today's class,
we are going to discuss about our project, Yoga post estimation
using ResNet. We are going to create our
project in the Google Colab. For that, you have to
create a new notebook. So we are going to use Google
Colab because Google Colab provides free GPU for
training our dataset. So after creating
the new notebook, you have to type
your project name. So I'm going to type
my project name, Yoga post deduction or
estimation using ResNet. So after typing
your project name, you have to change your runtime. So you have to
select T four GPU, and then you have
to select Save. So now we are going to discuss about what are all
the packages we need to import for our project Yoga post
detection using ResNet. Also, I'm going to use
my Google Drive because the Google Drive contains all
the dataset for my project. For that only I'm going to
import the Google Drive. So after that, you can see our Google Drive is
successfully connected. Now we are going to
import packages. So first, we need to
import a pandas package. So after that, we
are going to import Napi then we are going to
import fast AI package. So you have to type from
fast dot isen dot all input or we are going to input all the packages
from the fast AI. So it will be very useful
for training our dataset. So after that, we are going
to input the path package, so it will be very useful for providing the path
for our dataset. So these are all the packages. We need to input for our
Yoga post Deduction project. How to run the code so you can see all the packages
are successfully imported. So in the next class, we are going to discuss
about our dataset.
3. Yoga Class 2 : Import Dataset: Welcome, everyone.
In today's class, we are going to discuss
about our dataset for our project Yoga
post Detection. For that only, I have imported Google Drive because Google
Drive contains my dataset. So you can see my
dataset, Yoga post. I have created three dataset
yoga poses like Danda Sana, PTMA Sana, and VajraSNA. So these are all the yoga poses. So each folder contains the images so you can see
the images for PTMA Sana. So you can see all the images for every Yoga post deduction. So we're going to
use the dataset. So we need to import
the dataset for the Dototype dataset equaltPath
of inside the quotation, you have to copy the
path of your dataset, and then you have to paste the
path inside the quotation. So you have to run the code. So our path successfully
imported our dataset. So the next class
we are going to discuss about image
classification.
4. Yoga Class 3 : Classify Images For Train: Welcome, everyone.
In today's class, we are going to discuss
about image classification. So for that, we have to create
a variable data blocks. Data block, equal t, you have to type,
Data block off. So inside that we are going
to classify the image. First, we need to create blocks. We have to create two blocks, image block, and
then category block. So image block contains
all the images. Category block contains
three category. So the categories
like Danda SNA, PTMA Sana, and Wacha SNA. So these are all the
category blocks. After that, we need to extract all the images
from the folder. For that Yoo type, get items, equal to
get image files. It will extract all the
images from your folder. After that we need to
split the dataset for that oto type splitter,
equal, random splitter. Of valid equal to 0.2. So 20 percentage
separated for validation. And then remaining 80
percentage for training. So that's it. We successfully separated the dataset for
training and validation. So after that, we need to
provide label for our dataset. So you have to type get Y
equality, parent label. So labels are nothing
but the folder name. So the folders like Danda Sana, Patma Sana, and VajraSNA. So these are all the
labels for our dataset. So after that, we need to resize the images for our
training purpose. So you have to type IT TFS equal to resize of two to four. And then we are going to convert the folders to batches
for the Dototype batch, TF S equal to augment transforms of Multi Mult equality is 0.2. So it will create the batches. So we have successfully
classified our images into three categories Dandasana,
Butmasna and VajaSna. So these are all the yoga pose. You have to run the code. So in the next class, we are
going to discuss about how to train our dataset
using ResNet.
5. Yoga Class 4 : Train Dataset Using ResNet50: In today's class, we
are going to train our dataset using ResNet for our project yoga pose
estimation using Resont. For that, you need to
create a variable DLS equal to Data block
dot, data loads. We need to load our dataset. So you have to type
the variable dataset. So it contains the
path for our dataset. So after that, we have to find the batches what are all
the badges separated. For that only we need
to print the batches. You have to type DLS dot, so batch of maximum number NN batches will be
shown in the output. And then figure size, you have to type the height
and width Savankama six. You have to run the code, you have to wait
for the process. So you can see the batches. So these are all
the badges based on our folder name PTMA Sana, Danda Sana, and Vada sana. So before trading the dataset, you have to type the bats size. So you have to type bats
size equal to eight. So you can type
your own batch size based on your GPO performance, whether it can be 16, 32, 64 or 128. So after typing the batch size, we need to train the model
for that to type the variable learn equal Visen learner of you to type Visen
learner of DLS, Resonant 50, that's our model, and then metrics
equal to error rate. So if you want to
find the accuracy, you have to type accuracy
instead of error rate. So after that, we need to type the epoch for our
training purpose. You have to type learn dot, fine tool of you how to type the epoch inside
the quotation. You how to type 30. I'm going to use 30
epoch for the training. So how to run the code. So now we have to wait
for the training purpose. So you can say our model
is successfully trained. If you want to download
your trained model, you have to type the
variable, learn dot, export of inside the quotation, you can type your own
model file name dot PKL. So we are going to save
our format pickle file. So that's the format pickle PKL. So you can see the model in the Google Colab, yoga dot PKL. So that's our
trained model using the Resent 50. So that's it. We have successfully
trained our model. So in the next class, we are going to
complete our project.
6. Yoga Class 5 : Output & Conclusion: Everyone in today's class, we are going to complete
our project Yoga post deduction using ResNet. So we are going to predict
the images for the yoga pose, for that, you have to type the image path equal to
inside the quotation, you have to copy the
path of any image. I'm going to copy the
image from the Dandasana. And then I'm going to paste the path inside the quotation. Soft that I'm going to type, soft that I'm going to use my trained model for
prediction for that to to type image equal to plt mg dot
Create TAF image path. So we are going to use our trained model for
predicting the output. So you have to
predict the output. You have to type, predict, equal to learn, dot,
predict of image. We are going to
predict the yoga pose, and then you have to
print the prediction. You have to type
print of predict. So we are going to predict the
yoga pose to run the code, you can see our model is successfully predicted
the Yoga post DandaSana because we are copied the image from the
Danda Sana Yoga post. So that's why our model
source Dandasana. So if I select other images from Wada sana and then I'm
going to paste the image, and then I'm going to predict. You can see Vada Sana. Our model is well trained. So I'm going to
use the image from Batma Sana I'm going to
copy the image part, and then I'm going to paste the path and then run the file. So you can see Patm Sana,
successfully predicted. So if you want to see the image, you can use IMG dt so off. So it will sew the image in the output. So that's it, guys. We have successfully created our project Yoga post Deduction or prediction using Resent 50. Hope you learn something
from this course. So if you want to
download my code, it will be present
in the description. So let's see the next
course teached by me.
Arunnachalam Shanmugaraajan
