Best CNN architecture for binary classification of small images with a massive dataset [on hold]
1
$begingroup$
The title has it all...
Any tip is welcomed.
Should I use a very deep convolutional neural network ?
Use a large amount of filters ?
Parallel layers ?
Dataset examples:
1) "Good"
2) "Bad"
deep-learning dataset image-classification convolution accuracy
asked yesterday
Miko DikoMiko Diko
1062
New contributor
Miko Diko is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.
$endgroup$
put on hold as too broad by Icyblade, Siong Thye Goh, Mark.F, Kiritee Gak, Sean Owen♦ 10 hours ago
Please edit the question to limit it to a specific problem with enough detail to identify an adequate answer. Avoid asking multiple distinct questions at once. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.
add a comment |
1
$begingroup$
The title has it all...
Any tip is welcomed.
Should I use a very deep convolutional neural network ?
Use a large amount of filters ?
Parallel layers ?
Dataset examples:
1) "Good"
2) "Bad"
deep-learning dataset image-classification convolution accuracy
asked yesterday
Miko DikoMiko Diko
1062
New contributor
Miko Diko is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.
$endgroup$
put on hold as too broad by Icyblade, Siong Thye Goh, Mark.F, Kiritee Gak, Sean Owen♦ 10 hours ago
Please edit the question to limit it to a specific problem with enough detail to identify an adequate answer. Avoid asking multiple distinct questions at once. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.
$begingroup$
The dataset is made of 2 possible options: 1) An image like any other image you can think of..... 2) the image is "split" in the middle, the left part of the image was taken from 1 place, and the right side was taken from a different place...... so I want the model to tell "Continuous image, or 'cut' in the middle image".
$endgroup$
– Miko Diko
yesterday
$begingroup$
Can we see some example data please? Also a regular CNN would do fine with this.
$endgroup$
– JahKnows
yesterday
$begingroup$
posted examples
$endgroup$
– Miko Diko
18 hours ago
add a comment |
1
1
1
$begingroup$
The title has it all...
Any tip is welcomed.
Should I use a very deep convolutional neural network ?
Use a large amount of filters ?
Parallel layers ?
Dataset examples:
1) "Good"
2) "Bad"
deep-learning dataset image-classification convolution accuracy
asked yesterday
Miko DikoMiko Diko
1062
New contributor
Miko Diko is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.
$endgroup$
The title has it all...
Any tip is welcomed.
Should I use a very deep convolutional neural network ?
Use a large amount of filters ?
Parallel layers ?
Dataset examples:
1) "Good"
2) "Bad"
deep-learning dataset image-classification convolution accuracy
deep-learning dataset image-classification convolution accuracy
asked yesterday
Miko DikoMiko Diko
1062
New contributor
Miko Diko is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.
asked yesterday
Miko DikoMiko Diko
1062
New contributor
Miko Diko is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.
edited 18 hours ago
Miko Diko
asked yesterday
Miko DikoMiko Diko
1062
New contributor
Miko Diko is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.
asked yesterday
Miko DikoMiko Diko
1062
asked yesterday
Miko DikoMiko Diko
1062
1062
New contributor
Miko Diko is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.
New contributor
Miko Diko is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.
Miko Diko is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.
put on hold as too broad by Icyblade, Siong Thye Goh, Mark.F, Kiritee Gak, Sean Owen♦ 10 hours ago
Please edit the question to limit it to a specific problem with enough detail to identify an adequate answer. Avoid asking multiple distinct questions at once. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.
put on hold as too broad by Icyblade, Siong Thye Goh, Mark.F, Kiritee Gak, Sean Owen♦ 10 hours ago
Please edit the question to limit it to a specific problem with enough detail to identify an adequate answer. Avoid asking multiple distinct questions at once. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.
$begingroup$
The dataset is made of 2 possible options: 1) An image like any other image you can think of..... 2) the image is "split" in the middle, the left part of the image was taken from 1 place, and the right side was taken from a different place...... so I want the model to tell "Continuous image, or 'cut' in the middle image".
$endgroup$
– Miko Diko
yesterday
$begingroup$
Can we see some example data please? Also a regular CNN would do fine with this.
$endgroup$
– JahKnows
yesterday
$begingroup$
posted examples
$endgroup$
– Miko Diko
18 hours ago
add a comment |
$begingroup$
The dataset is made of 2 possible options: 1) An image like any other image you can think of..... 2) the image is "split" in the middle, the left part of the image was taken from 1 place, and the right side was taken from a different place...... so I want the model to tell "Continuous image, or 'cut' in the middle image".
$endgroup$
– Miko Diko
yesterday
$begingroup$
Can we see some example data please? Also a regular CNN would do fine with this.
$endgroup$
– JahKnows
yesterday
$begingroup$
posted examples
$endgroup$
– Miko Diko
18 hours ago
$begingroup$
The dataset is made of 2 possible options: 1) An image like any other image you can think of..... 2) the image is "split" in the middle, the left part of the image was taken from 1 place, and the right side was taken from a different place...... so I want the model to tell "Continuous image, or 'cut' in the middle image".
$endgroup$
– Miko Diko
yesterday
$begingroup$
The dataset is made of 2 possible options: 1) An image like any other image you can think of..... 2) the image is "split" in the middle, the left part of the image was taken from 1 place, and the right side was taken from a different place...... so I want the model to tell "Continuous image, or 'cut' in the middle image".
$endgroup$
– Miko Diko
yesterday
$begingroup$
Can we see some example data please? Also a regular CNN would do fine with this.
$endgroup$
– JahKnows
yesterday
$begingroup$
Can we see some example data please? Also a regular CNN would do fine with this.
$endgroup$
– JahKnows
yesterday
$begingroup$
posted examples
$endgroup$
– Miko Diko
18 hours ago
$begingroup$
posted examples
$endgroup$
– Miko Diko
18 hours ago
add a comment |
1 Answer
1
active
oldest
votes
0
$begingroup$
It all depends on the dataset, there is no single model can be the best.
I would prefer try a transfer learning Resnet34 or resnet50 with custom last layer for the number of class of classification.
answered yesterday
BenjiBBBenjiBB
387
$endgroup$
$begingroup$
Can't transfer learn. The dataset is made of the possible options: 1) An image like any other image you can think of..... 2) the image is "split" in the middle, the left part of the image was taken from 1 place, and the right side was taken from a different place...... so I want the model to tell "Continuous image, or 'cut' in the middle image".
$endgroup$
– Miko Diko
yesterday
$begingroup$
@MikoDiko in this case, I think a shallow convolution network with 3filters, stride one with one linear layer at the end would work well, it is very easy for a filter to detect the vertical split. I think you don’t even need a max pooling layer.
$endgroup$
– BenjiBB
yesterday
$begingroup$
Hi BenjiBB, that's what I did, I get ~85% accuracy, but how can I improve it to even 95% ? I tried 3 filters, withwithout max pooling, number of Fully-Connected layers, Batch Normalization .. what other trick there is ?
$endgroup$
– Miko Diko
18 hours ago
add a comment |
1 Answer
1
active
oldest
votes
1 Answer
1
active
oldest
votes
active
oldest
votes
active
oldest
votes
0
$begingroup$
It all depends on the dataset, there is no single model can be the best.
I would prefer try a transfer learning Resnet34 or resnet50 with custom last layer for the number of class of classification.
answered yesterday
BenjiBBBenjiBB
387
$endgroup$
$begingroup$
Can't transfer learn. The dataset is made of the possible options: 1) An image like any other image you can think of..... 2) the image is "split" in the middle, the left part of the image was taken from 1 place, and the right side was taken from a different place...... so I want the model to tell "Continuous image, or 'cut' in the middle image".
$endgroup$
– Miko Diko
yesterday
$begingroup$
@MikoDiko in this case, I think a shallow convolution network with 3filters, stride one with one linear layer at the end would work well, it is very easy for a filter to detect the vertical split. I think you don’t even need a max pooling layer.
$endgroup$
– BenjiBB
yesterday
$begingroup$
Hi BenjiBB, that's what I did, I get ~85% accuracy, but how can I improve it to even 95% ? I tried 3 filters, withwithout max pooling, number of Fully-Connected layers, Batch Normalization .. what other trick there is ?
$endgroup$
– Miko Diko
18 hours ago
add a comment |
0
$begingroup$
It all depends on the dataset, there is no single model can be the best.
I would prefer try a transfer learning Resnet34 or resnet50 with custom last layer for the number of class of classification.
answered yesterday
BenjiBBBenjiBB
387
$endgroup$
$begingroup$
Can't transfer learn. The dataset is made of the possible options: 1) An image like any other image you can think of..... 2) the image is "split" in the middle, the left part of the image was taken from 1 place, and the right side was taken from a different place...... so I want the model to tell "Continuous image, or 'cut' in the middle image".
$endgroup$
– Miko Diko
yesterday
$begingroup$
@MikoDiko in this case, I think a shallow convolution network with 3filters, stride one with one linear layer at the end would work well, it is very easy for a filter to detect the vertical split. I think you don’t even need a max pooling layer.
$endgroup$
– BenjiBB
yesterday
$begingroup$
Hi BenjiBB, that's what I did, I get ~85% accuracy, but how can I improve it to even 95% ? I tried 3 filters, withwithout max pooling, number of Fully-Connected layers, Batch Normalization .. what other trick there is ?
$endgroup$
– Miko Diko
18 hours ago
add a comment |
0
0
0
$begingroup$
It all depends on the dataset, there is no single model can be the best.
I would prefer try a transfer learning Resnet34 or resnet50 with custom last layer for the number of class of classification.
answered yesterday
BenjiBBBenjiBB
387
$endgroup$
It all depends on the dataset, there is no single model can be the best.
I would prefer try a transfer learning Resnet34 or resnet50 with custom last layer for the number of class of classification.
answered yesterday
BenjiBBBenjiBB
387
edited yesterday
answered yesterday
BenjiBBBenjiBB
387
answered yesterday
BenjiBBBenjiBB
387
answered yesterday
BenjiBBBenjiBB
387
387
$begingroup$
Can't transfer learn. The dataset is made of the possible options: 1) An image like any other image you can think of..... 2) the image is "split" in the middle, the left part of the image was taken from 1 place, and the right side was taken from a different place...... so I want the model to tell "Continuous image, or 'cut' in the middle image".
$endgroup$
– Miko Diko
yesterday
$begingroup$
@MikoDiko in this case, I think a shallow convolution network with 3filters, stride one with one linear layer at the end would work well, it is very easy for a filter to detect the vertical split. I think you don’t even need a max pooling layer.
$endgroup$
– BenjiBB
yesterday
$begingroup$
Hi BenjiBB, that's what I did, I get ~85% accuracy, but how can I improve it to even 95% ? I tried 3 filters, withwithout max pooling, number of Fully-Connected layers, Batch Normalization .. what other trick there is ?
$endgroup$
– Miko Diko
18 hours ago
add a comment |
$begingroup$
Can't transfer learn. The dataset is made of the possible options: 1) An image like any other image you can think of..... 2) the image is "split" in the middle, the left part of the image was taken from 1 place, and the right side was taken from a different place...... so I want the model to tell "Continuous image, or 'cut' in the middle image".
$endgroup$
– Miko Diko
yesterday
$begingroup$
@MikoDiko in this case, I think a shallow convolution network with 3filters, stride one with one linear layer at the end would work well, it is very easy for a filter to detect the vertical split. I think you don’t even need a max pooling layer.
$endgroup$
– BenjiBB
yesterday
$begingroup$
Hi BenjiBB, that's what I did, I get ~85% accuracy, but how can I improve it to even 95% ? I tried 3 filters, withwithout max pooling, number of Fully-Connected layers, Batch Normalization .. what other trick there is ?
$endgroup$
– Miko Diko
18 hours ago
$begingroup$
Can't transfer learn. The dataset is made of the possible options: 1) An image like any other image you can think of..... 2) the image is "split" in the middle, the left part of the image was taken from 1 place, and the right side was taken from a different place...... so I want the model to tell "Continuous image, or 'cut' in the middle image".
$endgroup$
– Miko Diko
yesterday
$begingroup$
Can't transfer learn. The dataset is made of the possible options: 1) An image like any other image you can think of..... 2) the image is "split" in the middle, the left part of the image was taken from 1 place, and the right side was taken from a different place...... so I want the model to tell "Continuous image, or 'cut' in the middle image".
$endgroup$
– Miko Diko
yesterday
$begingroup$
@MikoDiko in this case, I think a shallow convolution network with 3filters, stride one with one linear layer at the end would work well, it is very easy for a filter to detect the vertical split. I think you don’t even need a max pooling layer.
$endgroup$
– BenjiBB
yesterday
$begingroup$
@MikoDiko in this case, I think a shallow convolution network with 3filters, stride one with one linear layer at the end would work well, it is very easy for a filter to detect the vertical split. I think you don’t even need a max pooling layer.
$endgroup$
– BenjiBB
yesterday
$begingroup$
Hi BenjiBB, that's what I did, I get ~85% accuracy, but how can I improve it to even 95% ? I tried 3 filters, withwithout max pooling, number of Fully-Connected layers, Batch Normalization .. what other trick there is ?
$endgroup$
– Miko Diko
18 hours ago
$begingroup$
Hi BenjiBB, that's what I did, I get ~85% accuracy, but how can I improve it to even 95% ? I tried 3 filters, withwithout max pooling, number of Fully-Connected layers, Batch Normalization .. what other trick there is ?
$endgroup$
– Miko Diko
18 hours ago
add a comment |
$begingroup$
The dataset is made of 2 possible options: 1) An image like any other image you can think of..... 2) the image is "split" in the middle, the left part of the image was taken from 1 place, and the right side was taken from a different place...... so I want the model to tell "Continuous image, or 'cut' in the middle image".
$endgroup$
– Miko Diko
yesterday
$begingroup$
Can we see some example data please? Also a regular CNN would do fine with this.
$endgroup$
– JahKnows
yesterday
$begingroup$
posted examples
$endgroup$
– Miko Diko
18 hours ago