Will the proposed datasets be released ?

[Zeqiang-Lai]

1. Image Editing in the Wild (IEIW),
2. Object replacement based on SA-1B, Open Images
3. PhraseCut with LAMA
4. Web crawl.

[Gengzigang]

Thank you for your interest in the proposed datasets. Due to some copyright concerns, we currently have no plans to release these specific datasets. However, we believe that the publicly available datasets should be sufficient for training a model with similar performance to ours. We appreciate your understanding.

[Zeqiang-Lai]

Thanks for the rely, could you share the gpu budgets for constructing these datasets ?

[Gengzigang]

If you would like to train on all tasks (955k images per epoch) simultaneously, it took us 3.5 days to train for 200 epochs using 48 V100 GPUs. However, if removing IEIW or focusing on training specific tasks or datasets, our method is fully adaptable, and the training time will be significantly reduced.