allenai/Holodeck

Share Experiment Prompts or Selections in Paper

Closed this issue · 4 comments

Dear authors,

Thanks for the wonderful paper!

I would like to compare your method with my layout generation method, and in order to make a fair comparison I would like to use the prompts and better even, also the retrieved assets and room shapes that were used before generating the layout scene graphs with GPT-4.

Which experiment do you want to compare with? Could you please point me to the sections of the paper?

Here contains all the files from different systems in the experiment of Table 2.

To perform your own layout design module and keep assets the same, you refer to this function:

def ablate_placement(self, scene, query, save_dir, used_assets=[], add_ceiling=False, generate_image=True, generate_video=False, add_time=True, use_constraint=False, constraint_type="llm"):

and use your layout module to generate the position for all objects:
scene["floor_objects"] = self.floor_object_generator.generate_objects(scene, use_constraint=use_constraint)

Perfect! Thanks a lot for sharing :)