Diffusion models are initially designed for image generation. Recent research shows that the internal signals within their backbones, named activations, can also serve as dense features for various discriminative tasks such as semantic segmentation. Given numerous activations, selecting a small yet effective subset poses a fundamental problem. To this end, the early study of this field performs a large-scale quantitative comparison of the discriminative ability of the activations. However, we find that many potential activations have not been evaluated, such as the queries and keys used to compute attention scores.

We compare our method with four competing methods in Table 1 of the main paper. We also use the score reported by [5]. We found that the corner-based methods, e.g., HEA T and RoomFormer, fail to reconstruct the correct floorplans and are easily affected by the irregular Heat: Holistic edge attention transformer for structured reconstruction. Real-world perception for embodied agents.

Omnidirectional (or 360-degree) images are increasingly being used for 3D applications since they allow the rendering of an entire scene with a single image.

We now report the implementation details for the reproducibility of our method.