update

README.md

@@ -24,7 +24,7 @@ This benchmark consists of a subset of random **600 samples** from the **ShapeTa
 The repository is structured to seamlessly plug into the `Prox-E` unified evaluator:
 
 - **`input_shapes/`**: Directory containing `600` input 3D meshes in OBJ format (`.obj`).
-- **`rendered_images/`**: Directory containing `600` high-resolution upright 2D renders of the input meshes in PNG format (`.png`).
+- **`rendered_images/`**: 600 high-resolution upright 2D renders (PNG) of the input shapes, produced from TRELLIS's inversion outputs.
 - **`point_cloud/`**: Directory containing `600` dense point cloud of the input meshes (`.npz`).
 - **`instructions.json`**: Key-value metadata containing evaluation prompts, object classes and part-specific keywords.
 - **`shapetalk_600_set.csv`**: Full detailed metadata from the original ShapeTalk dataset on this 600-sample subset.