search

ai-med / relaynet_pytorch

Pytorch Implementation of retinal OCT Layer Segmentation (with trained models)
MIT License
87 stars 30 forks source link

The input data #2

Closed hwei-hw closed 6 years ago

hwei-hw commented 6 years ago

@abhi4ssj Could you open the 'Data.h5'? Thanks very much!!

zhewang95 commented 6 years ago

.mat data can be downloaded from http://people.duke.edu/~sf59/Chiu_BOE_2014_dataset.htm

hwei-hw commented 6 years ago

@zhewang95 Thanks very much for your reply! But the data from http://people.duke.edu/~sf59/Chiu_BOE_2014_dataset.htm is different from the author's ones. for examples, the rows of image is 512 in author's code but the image's rows is 496 in mat(Sub_1.mat, the image's size is [496，768，61]). So I don't know the author how to get the data?

hwei-hw commented 6 years ago

whether the trained PyTorch model can deal the image with different shape with train image ? for example, the train image's shape is [512,64,1] and the test image is [496,64,1]. is it right ?

zhewang95 commented 6 years ago

I think yes, the paper also say that.

hwei-hw commented 6 years ago

@zhewang95 Thank you very much! I will try it as soon as possible. Is it convenient to ask about your research direction? OCT image segmentation？if yes, I think we can communicate with each other if you do not mind.

zhewang95 commented 6 years ago

Sure, I'd like to. can't see your email, can you email me your wechat or what you like?

hwei-hw commented 6 years ago

@zhewang95 Thanks very much ! I will send a email for my WeChat

abhi4ssj commented 6 years ago

@Atomwh As the network is fully convolutional, it can handle test data of any size. I changed the data size to 512 rows by padding (power of 2), which makes it easy to track the intermediate feature map size. Also note that the data slicing code is not yet implemented in this version. It is in the original code in my repo.

hwei-hw commented 6 years ago

@abhi4ssj Thanks very much for your reply ! I'm trying to test my data.