Interpretable Diffusion via Information Decomposition

Authors: Xianghao Kong, Ollie Liu, Han Li, Dani Yogatama, Greg Ver Steeg

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We exploit these new relations to measure the compositional understanding of diffusion models, to do unsupervised localization of objects in images, and to measure effects when selectively editing images through prompt interventions. All our experiments are performed with pre-trained latent space diffusion models Rombach et al. (2022), Stable Diffusion v2.1 from Hugging Face unless otherwise noted.
Researcher Affiliation Academia Xianghao Kong1 * , Ollie Liu2 *, Han Li1, Dani Yogatama2, Greg Ver Steeg1 1University of California Riverside, 2University of Southern California {xkong016,hli358,gregoryv}@ucr.edu, {zliu2898, yogatama}@usc.edu
Pseudocode No The paper does not contain any clearly labeled "Pseudocode" or "Algorithm" blocks, nor does it present structured steps in a code-like format.
Open Source Code Yes See D for experiment details and links to open source code. Our code for reproducing the results is publicly available at: github.com/xkong016/interpretable_diffusion
Open Datasets Yes We start (left) with a real image from the COCO dataset. (Figure 1) ... we carefully filtered two datasets, COCO-IT and COCO-WL from the MSCOCO (Lin et al., 2015) validation dataset. ... we apply our pointwise estimator to analyze compositional understanding of Stable Diffusion on the ARO benchmark Yuksekgonul et al. (2022).
Dataset Splits Yes we carefully filtered two datasets, COCO-IT and COCO-WL from the MSCOCO (Lin et al., 2015) validation dataset. (Section 3.2) ... We follow Yuksekgonul et al. (2022) to construct the validation sets for VG-A, VG-R, COCO and Flickr30k. (Appendix D.1)
Hardware Specification No The paper does not provide specific hardware details such as GPU/CPU models, processor types, or memory amounts used for running its experiments.
Software Dependencies Yes All our experiments are performed with pre-trained latent space diffusion models Rombach et al. (2022), Stable Diffusion v2.1 from Hugging Face unless otherwise noted.
Experiment Setup Yes We use a truncated logistic for the importance sampling distribution for α. (Section 2.4) ... The solver we use for this ODE is the 2nd order deterministic solver with 100 steps from Karras et al. (2022). (Section 3.3) ... Table 3 lists experiment configurations with '50 steps', '100 steps', '200 steps'.