How to make data augmentation

irolaina / FCRN-DepthPrediction

Deeper Depth Prediction with Fully Convolutional Residual Networks (FCRN)

BSD 2-Clause "Simplified" License

1.11k stars 313 forks source link

How to make data augmentation #37

Closed E-MHJ closed 6 years ago

E-MHJ commented 6 years ago

Hi ,I am trying to recreate your results on the NYU_V2 dataset，but the huber loss can't converge when training. I guess i need to make data augmentation, but i am unsure how to do it. Now i have 12k image pairs, if i want to get 95K pairs, should i make rotation, scaling, color transformations and flip eight times for each RGB-D image pair ? Besides, when computing loss, if the input groundtruth depth size is 640x480? Thank you.

Ariel-JUAN commented 6 years ago

@E-MHJ you need data augmentation, following as Elign's paper. Besides, can you show me you loss code? My loss converge but the result is not very good.

mortalsoulzy commented 6 years ago

@Ariel-JUAN Hi,I'm trying to reproduce this experiment with tensorflow , but after training for few steps ,the output depth map convergence to the same number for all pixels.And don't change anymore.I'm quite confused for that,can u share some training details with me?That would be of great help,Tanks~

Ariel-JUAN commented 6 years ago

I don't really understand your question...... maybe you can leave you email and I will contact you.

mortalsoulzy commented 6 years ago

@Ariel-JUAN My email address is mortalsoulzy@gmail.com，thank u so much~