Deep Mean-Shift Priors for Image Restoration

Authors: Siavash Arjomand Bigdeli, Matthias Zwicker, Paolo Favaro, Meiguang Jin

NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5 Experiments and Results Our DAE uses the neural network architecture by Zhang et al. [39]. We generated training samples by adding Gaussian noise to images from Image Net [10]. We experimented with different noise levels and found σ1 = 11 to perform well for all our deblurring and super-resolution experiments. Unless mentioned, for image restoration we always take 300 iterations with step length α = 0.1 and momentum µ = 0.9.
Researcher Affiliation Academia Siavash A. Bigdeli University of Bern bigdeli@inf.unibe.ch Meiguang Jin University of Bern jin@inf.unibe.ch Paolo Favaro University of Bern favaro@inf.unibe.ch Matthias Zwicker University of Bern, and University of Maryland, College Park zwicker@cs.umd.edu
Pseudocode Yes Table 1: Gradient descent steps for non-blind (NB), noise-blind (NA), and kernel-blind (KE) image deblurring.
Open Source Code Yes 1The source code of the proposed method is available at https://github.com/siavashbigdeli/DMSP.
Open Datasets Yes We generated training samples by adding Gaussian noise to images from Image Net [10].Table 2 reports the average PSNR for 32 images from the Levin et al. [19] and 50 images from the Berkeley [2] segmentation dataset
Dataset Splits No No explicit details on validation set splits (percentages, counts, or specific pre-defined splits) are provided in the paper.
Hardware Specification Yes The runtime of our method is linear in the number of pixels, and our implementation takes about 0.2 seconds per iteration for one megapixel on an Nvidia Titan X (Pascal).
Software Dependencies No The paper mentions 'Our DAE uses the neural network architecture by Zhang et al. [39]' but does not provide specific software dependency names with version numbers (e.g., Python, PyTorch, TensorFlow, CUDA versions).
Experiment Setup Yes Unless mentioned, for image restoration we always take 300 iterations with step length α = 0.1 and momentum µ = 0.9.We used momentum µ = 0.7 and step size α = 0.3 for the unknown image and momentum µk = 0.995 and step size αk = 0.005 for the unknown kernel.