Optimize & Reduce: A Top-Down Approach for Image Vectorization

Authors: Or Hirschorn, Amir Jevnisek, Shai Avidan

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

Reproducibility Variable Result LLM Response
Research Type Experimental Through extensive experiments on hundreds of images, we demonstrate that our method is domain agnostic and outperforms existing works in both reconstruction and perceptual quality for a fixed number of shapes.
Researcher Affiliation Academia Tel-Aviv University, Israel orhirschorn@mail.tau.ac.il, amirjevn@mail.tau.ac.il, avidan@eng.tau.ac.il
Pseudocode No No structured pseudocode or algorithm blocks were found. The method is described in prose and illustrated with diagrams (Figures 3 and 4) and mathematical equations.
Open Source Code Yes Our code is publicly available: https://github.com/ajevnisek/optimize-and-reduce
Open Datasets Yes For evaluation, we collect five datasets to cover a range of image complexities... All datasets will be made publicly available. EMOJI dataset. We take two snapshots of the Noto Emoji project (Google 2022).
Dataset Splits No The paper does not provide specific dataset split information (e.g., percentages or exact counts for training, validation, and test sets). It mentions the datasets used and general experimental settings but not the data partitioning methodology.
Hardware Specification Yes Runtime was measured on an NVIDIA RTX A5000 GPU.
Software Dependencies No The paper states, 'Our code is written in Py Torch (Paszke et al. 2017),' but does not provide a specific version number for PyTorch or any other software dependencies crucial for reproduction.
Experiment Setup Yes We use three Reduce steps that follow an exponential decay schedule (i.e., the number of shapes is halved in each iteration). For the qualitative experiments, we use MSE for optimization and L1 for reduction. All quantitative evaluations use L1 with our geometric loss for optimization. For reduction, we use L1 and Clip on Midjourney, and for the other datasets, we use only L1.