search

raywzy / ICT

High-Fidelity Pluralistic Image Completion with Transformers (ICCV 2021)
313 stars 44 forks source link

Training Data Format #19

Open AhmedHessuin opened 3 years ago

AhmedHessuin commented 3 years ago

can you tell me the training data format? as i understand 1) training data contain images for the pretrained models 256 x 256 in directory for example X 2) training data contain the masks of the images same size as the original train image for my example 256x256 in directory Y 3) optional, if i used flag --random_stroke, i don't need the training data contain the masks if i am right, please confirm and if i am wrong please tell me the right

thanks in advance

raywzy commented 3 years ago

Sorry for the late reply due to other ddl. Basically, yes. Feel free to raise any other questions further.

administratorssssssss commented 2 years ago
  1. Use the 1024 of the FFHQ dataset to reset to 256X256 size of the image as Y
  2. The 256X256 size picture is the input X after adding MASK? What is the format of the data, and can you send me some sample pictures? thanks！