Scribd
Upload
58 views2 pages

Gesture Recognition - WriteUp

The document describes 10 experiments conducted to develop a gesture recognition model using 3D convolutional neural networks. The first experiment achieved an accuracy of 50% and loss of 1.433, outperforming subsequent experiments that tested different model configurations, including changes to regularization parameters, frame size, batch size, and image size. The conclusion is that the initial model with a 180x180 image size, batch size of 32 or 64, and 4 CNN layers performed best without overfitting.

Uploaded by

Vaishak Jayaram
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
58 views2 pages

Gesture Recognition - WriteUp

The document describes 10 experiments conducted to develop a gesture recognition model using 3D convolutional neural networks. The first experiment achieved an accuracy of 50% and loss of 1.433, outperforming subsequent experiments that tested different model configurations, including changes to regularization parameters, frame size, batch size, and image size. The conclusion is that the initial model with a 180x180 image size, batch size of 32 or 64, and 4 CNN layers performed best without overfitting.

Uploaded by

Vaishak Jayaram
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Gesture Recognition

Experiment Model Metrics Decision +


Number Explanation

1 Conv3D Accuracy : 50% Tried a basic model,


using Adam
Loss : 1.433 Optimiser having a
batch size of 32.
The model was
overfitting the data.

2 Conv3D Accuracy : 18.75% Added regularisation


parameters , here
Loss : 5.44 we saw a drop in
accuracy and the
model was still
overfitting

3 Conv3D Accuracy : 37.5% Added more


regularisation
Loss : 1.51 parameters , the
accuracy had nearly
doubled from the
previous model but
isn’t better than our
first model.

4 Conv3D Accuracy : 31.5% Changing frame size


from 10 to 20 is
Loss : 2.64 impacting the
training time but we
are not getting good
accuracy.

5 Conv3D Accuracy : 31% Changing frame


from 20 to 30. The
Loss : 4.83 previous experiment
gave us almost
similar accuracy
with a lower loss. So
we’ll stick to 20
frame size.

6 Conv3D Accuracy : 50% Changing Batch


Loss : 1.76 Size to 64. This has
not brought any
change to the first
experiment.
7 Conv3D Accuracy : 43.75% Changing Image
Size (360*360) and
Loss : 3.082 Batch Size ( 32 ).

8 Conv3D Accuracy : 31% Changing Image


Size (360*360) and
Loss : 3.40 Batch Size ( 64 ).
The Hardware is not
able to handle the
image size and
batch size

9 Conv3D Accuracy : 50% Reducing the


network complexity
Loss : 1.54 since above models
are overfitting.

10 Conv3D Accuracy : 43.75% Increase further


complexity of the
model with more
Loss : 1.186 number of layers.
also, more
regularization is
added to check the
overfitting

Conclusion :

Overall the first experiment has the better metrics compared to the others and we
can also say that models with an image size of 180*180 having batch size 32 or 64 , with 4
CNN layer are working better.

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy