How we can use it ? about street-view-house-numbers-svhn-detection-and-classification-using-cnn

[pavitrakumar78]

Not sure what you mean, but if you are asking what improvement can be made, here are some things you can do:

• Come up with better CNN architectures
• Try to use better pre-processing methods
• Improve the bounding box estimation logic
• Try using other images and custom labeled data as input

These are just high-level ideas, but you can work on improving even small parts of each section which may improve accuracy even by a little bit.

[Malouke]

hello ,
thank you again for reply,
I mean i want test your project in personal images but i am not expert and i want realy some basics stuffs like how i can use your model to try detect numbers in my personnal pictures.

How i can launch the script ?
where i should put my images ?
do i should train the modele or use the save weights in your google drive .... ?

thanks for adance .

[pavitrakumar78]

All the required instructions are on the readme itself. Download the full repo and put the weights and data files into the respective folders and run combi_models.py.
Whether you want to train or use pre-trained models - that's up to you.

[Malouke]

ok thank 's for your reply,
i am so sorry if my question was no clear.

my simple question it's where i should put my own images for testing ?

thank's for adavance.

[pavitrakumar78]

Anywhere you want. Read the code here. find_box_and_predict_digit() is what you need and it's input is a numpy array. You have to read the image and pass it to the function. Do look at how the images are processed for training and make sure you do the same processing if you want to use the pre-trained models.

[Malouke]

thank's for your help .
i test he work well ,
i test only on my cpu i want ask you if i can use my gpu GTX 2 GB memory to play with data ?

[pavitrakumar78]

Yes, you can use your GPU.

[Malouke]

thank's again