search

V-Sense / DeepNormals

Code and Dataset from Deep Normal Estimation for Automatic Shading of Hand-Drawn Characters
Other
181 stars 18 forks source link

Training an Image #6

Closed JuHyung-Son closed 5 years ago

JuHyung-Son commented 5 years ago

When training, did you train every tile of an image among 420 images or just a random tile among 84000 tiles?

MatisHudon commented 5 years ago

If I understand your question correctly: Random tile among 84000, we generated a tfrecord file containing all tiles (20 tiles per image), then used the shuffle batch during training. Contact me by email if you need more info or the tfrecord file directly. (hudonm@scss.tcd.ie)