Isometric Representation Learning for Disentangled Latent Space of Diffusion Models

Authors: Jaehoon Hahm, Junho Lee, Sunghyun Kim, Joonseok Lee

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments illustrate advantages of the proposed method in image interpolation, image inversion, and linear editing. and We conduct extensive experiments to verify the effectiveness of our isometric loss Liso on disentangling the latent space of diffusion models.
Researcher Affiliation Collaboration 1Seoul National University, Seoul, Korea 2Google Research, Mountain View, California, United States.
Pseudocode No The paper does not contain any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes Our source code is publicly available at https://github.com/isno0907/isodiff. Readers would be able to reproduce the reported results by running this code.
Open Datasets Yes Dataset. We evaluate our approach on CIFAR-10, Celeb AHQ (Huang et al., 2018), LSUN-Church and LSUNBedrooms (Wang et al., 2017).
Dataset Splits No The paper mentions training partition sizes for each dataset, but does not explicitly provide details about train/validation/test splits (percentages or counts) that would allow reproduction of the data partitioning.
Hardware Specification Yes We use 4 NVIDIA A100 GPUs with 40GB memory for experiments.
Software Dependencies No The paper mentions the use of an Adam optimizer and U-Net architecture, but does not specify software dependencies with version numbers (e.g., Python, PyTorch/TensorFlow versions).
Experiment Setup Yes If not specified, we train with batch size 32, learning rate 10 4, p = 0.5, and λiso = 10 4 for 10 epochs by default. ... We use Adam optimizer and exponential moving average (Brown, 1956) on model parameters with a decay factor of 0.9999.