Progress

1. More training epochs on Dunhuang Cave painting generations

   1. link:

      stylegan2使用敦煌壁画作为训练集训练

   2. 56+80+32=168 ticks in total

   3. analysis

   • example: generated fake murals
2. Develop a Dunhuang portrait dataset with 902 images
   1. python work to do auto face recognition and alignment
   2. python work for os file stuff + Sketch manual clip and alignment
   • screenshot of the manual work in software Sketch
3. Train a styleGAN2 model for generating Dunhuang face/portrait.

   1. link:

      stylegan2使用敦煌壁画人脸作为训练集训练

   2. 288+424 = 712 ticks in total

   • example: generated fake portraits during training
4. Train a U-GAT-IT model for "human face to Dunhuang portrait" translation

   1. link:

      U-GAT-IT使用敦煌人脸进行训练

   2. 33 epochs till now, it is still training

5. Input an Asian face dataset to train the U-GAT-IT
   1. link:
   2. 2 epochs till now, it is still training
   3. updated in November 9th: 88 epochs with good transferred results
   4. updated in November 16th, both reached 100 epochs and finished the training.

Paper Outline

Paper-Outline/Drafts

Questions

• About evaluation
• I may also add more literature and art, perspectives and contexts to support my purpose of protecting Dunhuang art & culture heritage, as well as the data bias.