Ultrafast Photorealistic Style Transfer via Neural Architecture Search

Authors: Jie An, Haoyi Xiong, Jun Huan, Jiebo Luo10443-10450

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive experiments on both image and video transfer. The results show that our method can produce favorable results while achieving 20-30 times acceleration in comparison with the existing state-of-the-art approaches.
Researcher Affiliation Collaboration Jie An, 1 Haoyi Xiong, 2 Jun Huan,3 Jiebo Luo1 1University of Rochester, 2Baidu Research, 3Styling AI Inc.
Pseudocode No The paper describes the proposed method using figures and prose, but does not include any formal pseudocode or algorithm blocks.
Open Source Code No All the source code will be made released in the future.
Open Datasets Yes The decoder (without transfer modules) is trained on MS COCO dataset (Lin et al. 2014) to invert deep features of the encoder back to images.
Dataset Splits Yes Given the MS COCO as the training dataset and a validation dataset with 40 content and style photos, we first train Photo Net as the Supervisory Oracle for the subsequent architecture search. ... Given a validation dataset contains 73 content and style photo pairs, we quantitatively evaluate the performance of the proposed and state-of-the-art methods by computing the above-mentioned metrics on this validation set.
Hardware Specification Yes All approaches are tested on the same computing platform which includes an NVIDIA P100 GPU card with 16GB RAM.
Software Dependencies No The paper does not provide specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup No The paper mentions that the encoder is fixed and the decoder is trained for image reconstruction, and that hyperparameters (α, β, γ) are used for trade-off in the search objective, but it does not provide their specific values or other detailed hyperparameters like learning rate, batch size, or optimizer settings.