- Put all meshes under
inputs/(e.g.,inputs/dolphin.obj,inputs/teapot.obj,inputs/Tree.obj). - All rendered images are written to
outputs/. Each script will create this folder if it does not exist.
-
run.shruns our method on the 3 benchmark objects (dolphin, teapot, tree) and saves:outputs/dolphin_ours.pngoutputs/teapot_ours.pngoutputs/tree_ours.png
-
run_bas.shruns the CLIP-only baseline on the same 3 objects and saves:outputs/dolphin_clip.pngoutputs/teapot_clip.pngoutputs/tree_clip.png
These two scripts together produce the benchmark comparison figure inputs (top row: ours, bottom row: CLIP-only) using the same target prompts per object.
Run:
bash run.sh
bash run_bas.sh-
run_com.shruns two prompts on the teapot only, saving:outputs/teapot_con.png(concrete prompt)outputs/teapot_abs.png(abstract prompt)
Run:
bash run_com.sh-
run_ext.shruns three progressively longer prompts on the teapot only, saving:outputs/teapot_ext1.pngoutputs/teapot_ext2.pngoutputs/teapot_ext3.png
Run:
bash run_ext.sh-
Our method outputs are named
*_ours.png. -
Baseline outputs are named
*_clip.png. -
Ablation outputs follow:
- Concrete vs Abstract:
*_con.png,*_abs.png. - Length extension:
*_ext1.pngto*_ext3.png.
- Concrete vs Abstract: