Cold Diffusion: Inverting Arbitrary Image Transforms Without Noise
Authors: Arpit Bansal, Eitan Borgnia, Hong-Min Chu, Jie Li, Hamid Kazemi, Furong Huang, Micah Goldblum, Jonas Geiping, Tom Goldstein
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We emprically evaluate generalized diffusion models trained on different degradations with TACo S proposed in Algorithm 2. We perform experiments on the vision tasks of deblurring, inpainting, and super-resolution. We perform our experiments on MNIST [Le Cun et al., 1998], CIFAR-10 [Krizhevsky, 2009], and Celeb A [Liu et al., 2015]. In each of these tasks, we gradually remove the information from the clean image, creating a sequence of images such that D(x0, t) retains less information than D(x0, t 1). For these different tasks, we present both qualitative and quantitative results on a held-out testing dataset and demonstrate the importance of the sampling technique described in Algorithm 2. For all quantitative results in this section, the Frechet inception distance (FID) scores [Heusel et al., 2017] for degraded and reconstructed images are measured with respect to the testing data. |
| Researcher Affiliation | Academia | 1University of Maryland 2New York University |
| Pseudocode | Yes | Algorithm 1 Naive Sampling (Eg. DDIM) Input: A degraded sample xt for s = t, t 1, . . . , 1 do ˆx0 R(xs, s) xs 1 = D(ˆx0, s 1) end for Return: x0 |
| Open Source Code | No | The paper does not provide an explicit statement about releasing the source code for its methodology or a link to a code repository. |
| Open Datasets | Yes | We perform our experiments on MNIST [Le Cun et al., 1998], CIFAR-10 [Krizhevsky, 2009], and Celeb A [Liu et al., 2015]. |
| Dataset Splits | No | The paper mentions 'held-out validation set' in the context of FID scores for inpainting, but does not provide specific details on the dataset splits (percentages, sample counts, or explicit splitting methodology) for training, validation, and testing needed for reproduction. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, or memory amounts) used for running its experiments. |
| Software Dependencies | No | The paper mentions using the 'Adam Kingma and Ba [2014] optimizer' but does not specify versions for any key software components or libraries (e.g., Python, PyTorch, CUDA) needed to replicate the experiment. |
| Experiment Setup | Yes | We use the Adam Kingma and Ba [2014] optimizer with learning rate 2 10 5. The training was done on the batch size of 32, and we accumulate the gradients every 2 steps. Our final model is an Exponential Moving Average of the trained model with decay rate 0.995 which is updated after every 10 gradient steps. |