search

xinwucwp / faultSeg

Using synthetic datasets to train an end-to-end CNN for 3D fault segmentation (We are working on an improved version!)
223 stars 136 forks source link

Prediction Output images look different #11

Open ravi9 opened 4 years ago

ravi9 commented 4 years ago

Hi Dr Wu, I reran the prediction.ipynb, but I get different images compared to your output. Could you kindly advise if there is anything missing. Thank for your time @xinwucwp Please see my output here:https://github.com/ravi9/faultSeg/blob/master/prediction.ipynb

compare-outputs

Thanks, Ravi.

llmpass commented 2 years ago

Hi Ravi,

Your inputs shown in the right image look very different to the images shown in the left. My guess is that you didn't perform normalization of the data before feeding into the network.

yours, Luming

ravi9 commented 2 years ago

Hi @llmpass , Thank you for reviewing this. I'm using the same dataset as listed in the .ipynb and same code. Not sure what kind of additional normalization is needed.

I'm guessing the dataset is updated. Because, The predNew.ipynb shows the images in a similar way. They look similar to my images (right side).

Regards, Ravi.

llmpass commented 2 years ago

try minus mean and divide by std