Dear TRELLIS.2 authors,
First of all, thank you so much for your outstanding work and for open-sourcing the code! TRELLIS.2 is truly impressive, and we deeply appreciate the effort you have put into both the research and the release.
We noticed that in the paper, you mention using 100 AI-generated image prompts (reference [16]) for generation quality comparison and user studies, specifically to ensure training–testing disjointness. We are very interested in reproducing the evaluation results reported in the paper and conducting fair comparisons with other methods.
Could you kindly consider releasing any of the following:
The 100 AI-generated images used as input prompts in your evaluation pipeline, or
The text prompts used to generate those 100 images (so that the community can regenerate them using the same or similar tools), or
A pointer to the original source (reference [16]) where these images can be obtained, along with any preprocessing steps applied.
We understand that releasing evaluation assets may involve additional effort, and we completely respect your timeline and priorities. Even a brief clarification on whether this is planned for future release would be greatly appreciated.
Thank you again for your wonderful contribution to the community. We look forward to hearing from you!
Best regards,
Haoshu Wu
Dear TRELLIS.2 authors,
First of all, thank you so much for your outstanding work and for open-sourcing the code! TRELLIS.2 is truly impressive, and we deeply appreciate the effort you have put into both the research and the release.
We noticed that in the paper, you mention using 100 AI-generated image prompts (reference [16]) for generation quality comparison and user studies, specifically to ensure training–testing disjointness. We are very interested in reproducing the evaluation results reported in the paper and conducting fair comparisons with other methods.
Could you kindly consider releasing any of the following:
The 100 AI-generated images used as input prompts in your evaluation pipeline, or
The text prompts used to generate those 100 images (so that the community can regenerate them using the same or similar tools), or
A pointer to the original source (reference [16]) where these images can be obtained, along with any preprocessing steps applied.
We understand that releasing evaluation assets may involve additional effort, and we completely respect your timeline and priorities. Even a brief clarification on whether this is planned for future release would be greatly appreciated.
Thank you again for your wonderful contribution to the community. We look forward to hearing from you!
Best regards,
Haoshu Wu