search

fenglinglwb / MAT

MAT: Mask-Aware Transformer for Large Hole Image Inpainting
Other
728 stars 83 forks source link

image paint #87

Open lijain opened 1 year ago

lijain commented 1 year ago

Hello, may I ask why the eyes and mouths of some animals are somewhat distorted when I perform finetune on my data set and use a small mask to repair images of the same size as 512? Is there a problem with my Settings? image

image

fenglinglwb commented 1 year ago

The model pre-trained on Places sees scenes rather than individual objects. You could prolong the fine-tuning process.

lijain commented 1 year ago

Thank you for your patient answer. There are many types of data in this section, not only scenes but also animals. Data 60w I've fine-tuned 17 rounds

lijain commented 1 year ago

Hello, I wonder if you can explain? Is it because animals have different eyes than humans?

fenglinglwb commented 1 year ago

Sure, there is almost no such data in Places. You may train from scratch to see whether this helps.

sofpya commented 4 months ago

Hello, I am also trying to complement the animal's face using this code. Have you solved this problme?