SeeClear: Semantic Distillation Enhances Pixel Condensation for Video Super-Resolution

Authors: Qi Tang, Yao Zhao, Meiqin Liu, Chao Yao

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

Reproducibility Variable Result LLM Response
Research Type Experimental Comprehensive experiments confirm our framework s advantage over state-of-the-art diffusion-based VSR techniques.
Researcher Affiliation Academia Qi Tang1,2, Yao Zhao1,2, Meiqin Liu1,2 , Chao Yao3 1 Institute of Information Science, Beijing Jiaotong University 2 Visual Intelligence + X International Cooperation Joint Laboratory of MOE, Beijing Jiaotong University 3 School of Computer and Communication Engineering, University of Science and Technology Beijing {qitang, yzhao, mqliu}@bjtu.edu.cn, yaochao@ustb.edu.cn
Pseudocode Yes Algorithm 1: Generation Process of Slee Cear
Open Source Code Yes The code is available: https://github.com/Tang1705/See Clear-Neur IPS24.
Open Datasets Yes To assess the effectiveness of the proposed See Clear, we employ two commonly used datasets for training: REDS [24] and Vimeo-90K [41].
Dataset Splits Yes In accordance with the conventions established in previous works [1, 3], we select four clips2 from the training dataset to serve as a validation dataset, referred to as REDS4.
Hardware Specification Yes The See Clear framework is implemented with Py Torch-2.0 and trained across 4 NVIDIA 4090 GPUs, each accommodating 4 video clips.
Software Dependencies Yes The See Clear framework is implemented with Py Torch-2.0
Experiment Setup Yes All training stages utilize the Adam optimizer with β1 = 0.5 and β2 = 0.999, where the learning rate decays with the cosine annealing scheme. The Charbonnier loss [4] is applied on the whole frames between the ground truth and the reconstructed frame, formulated as L = p ||IHR i ISR i ||2 + ϵ2.