search

ChrisChen1023 / HINT

HINT: High-quality INpainting Transformer with Enhanced Attention and Mask-aware Encoding
MIT License
30 stars 4 forks source link

Number of training iterations #15

Closed mumingerlai closed 5 months ago

mumingerlai commented 5 months ago

Can I ask how many iterations of training you did? I noticed that your ablation experiment iterated 30,000 rounds and the results are different from the comparison experiment.

ChrisChen1023 commented 5 months ago

Hi,

The ablation study is trained for 30,000 iterations only. We fully trained our models for around two weeks on a single A100 GPU. Please follow our config and the performance should be same.

mumingerlai commented 5 months ago

Hi,

The ablation study is trained for 30,000 iterations only. We fully trained our models for around two weeks on a single A100 GPU. Please follow our config and the performance should be same.

Thank you very much for your reply. I checked your configuration and it seems that you trained for 300,000 iterations. Is the number of iterations the same for each dataset?

ChrisChen1023 commented 5 months ago

Hi, The ablation study is trained for 30,000 iterations only. We fully trained our models for around two weeks on a single A100 GPU. Please follow our config and the performance should be same.

Thank you very much for your reply. I checked your configuration and it seems that you trained for 300,000 iterations. Is the number of iterations the same for each dataset?

That 300,000 iterations is just for default setting. Please change the number for different datasets, depends on the complexity of datasets.