Please explain steps to do with datasets and training!

[tommyngx]

Dear @tusharsangam, Thank you for your sharing, I love your work. I would like to learn from your repo.

• I understand the dataset privacy, if I have a knee X-ray image, what should I do to make it the right format for the training?
• Could you explain the basic steps of training and testing? Thank you!

[tusharsangam]

@Tommy-Ngx Thank you for reaching out to me, It good to see that my work is being useful, I regret that I didn't document it well but I will be happy to help, So in the final I use only Normalize & ContrastAdjustment as my final transforms for Unet, Normalize changes the range of the image to -1 to 1, while ContrastAdjustment is not required but I have done it, most of my code is used from Integrating Spatial Configuration into Heatmap Regression Based {CNNs} for Landmark Localization so you can review it

Also if you don't have enough data I have generated an augmentation schedule based on SITK library which I found in the same repo as the paper I mentioned, I first replicate or augment the data save the image & label pair names in pickle file & then make a custom dataloader based on pickle file data This data loader is very basic & mimics basic functionality that you can find on pytorch's website tutorial

It's been almost 1 year I did this project & so even it's difficult for me to point out the exact so if you need further help we can arrange a zoom meeting & discuss what you are looking for

[tommyngx]

Thank you! I definitely spend more time and contact you if I have any questions.