Regarding the Training Dataset of UnifiedReward-Edit-qwen3vl

[wzh326]

The performance of UnifiedReward-Edit-qwen3vl exceeds expectations remarkably well. We are very eager to learn about the specific datasets you used for training, as well as the respective processing methods applied to each type of data. We would be extremely grateful if these datasets could be open-sourced—this would undoubtedly be an outstanding contribution!

[wzh326]

Also, it would be greatly appreciated if the author could provide details on the training methodology for UnifiedReward-Edit-qwen3vl. I am very interested in understanding how the text-to-image and image-to-image tasks are jointly trained in this framework.

[CodeGoat24]

Thank you for the feedback!

We will open-source the training data and methodology as soon as possible.