Boundary-aware Decoupled Flow Networks for Realistic Extreme Rescaling

Authors: Jinmin Li, Tao Dai, Jingyun Zhang, Kang Liu, Jun Wang, Shaoming Wang, Shu-Tao Xia, Rizen Guo

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

Reproducibility Variable Result LLM Response
Research Type Experimental Comprehensive experiments demonstrate that our BDFlow significantly outperforms other stateof-the-art methods while maintaining lower complexity.
Researcher Affiliation Collaboration Jinmin Li1, , Tao Dai2, , Jingyun Zhang4 , Kang Liu1 , Jun Wang4 , Shaoming Wang4 , Shu-Tao Xia1,3 , Rizen Guo4 1Tsinghua Shenzhen International Graduate School, Tsinghua University 2College of Computer Science and Software Engineering, Shenzhen University 3Research Center of Artificial Intelligence, Peng Cheng Laboratory 4We Chat Pay Lab33, Tencent
Pseudocode Yes Algorithm 1 Boundary-aware Mask Input: I, input image Parameter: T, the threshold to sparsify boundary Output: Bq, boundary distribution 1: Ig Gauss Blur(I, σ) 2: Gx, Gy Gradient(Ig) 3: M get Magnitude(Gx, Gy) 4: M Non Maximum Suppress(M) 5: Bs Sparsify Boundary(M , T) 6: Bq Quantify(Bs) 7: return Bq
Open Source Code Yes The code will be available at https://github.com/THU-Kingmin/BAFlow.
Open Datasets Yes We utilize the Celeb A-HQ dataset [Karras et al., 2020] to train our BDFlow. The dataset comprises 30,000 high-resolution (1024 1024) human face images. Moreover, we evaluate our models using widely-accepted pixel-wise metrics, including PSNR, SSIM, and LPIPS [Zhang et al., 2018a] (Y channel) on the Celeb A-HQ test dataset. We additionally train our BDFlow on the Cat dataset [Zhang et al., 2008] and the LSUN-Church [Yu et al., 2015] dataset to evaluate its generalization ability across different domains.
Dataset Splits No The paper mentions using Celeb A-HQ, Cat, and LSUN-Church datasets for training and a Celeb A-HQ test dataset for evaluation, but does not specify explicit train/validation/test splits (e.g., percentages, sample counts, or predefined split methods) for reproduction.
Hardware Specification No The paper does not provide specific hardware details such as GPU/CPU models, processor types, or memory amounts used for running experiments.
Software Dependencies No The paper mentions the ADAM optimizer but does not specify any software libraries or frameworks with version numbers (e.g., PyTorch 1.x, TensorFlow 2.x, CUDA 11.x) that are required to replicate the experiments.
Experiment Setup Yes We train these models using the ADAM [Kingma and Ba, 2014] optimizer with β1 = 0.9 and β2 = 0.999. Furthermore, we initialize the learning rate at 2 10 4 and apply a cosine annealing schedule, decaying from the initial value to 1 10 6 over the total number of iterations. We set λ1, λ2, λ3, λ4, and λ5 to 2, 2, 1, 16, and 4, respectively.