Blind Video Temporal Consistency via Deep Video Prior

Authors: Chenyang Lei, Yazhou Xing, Qifeng Chen

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

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the effectiveness of our approach on 7 computer vision tasks on videos. Extensive quantitative and perceptual experiments show that our approach obtains superior performance than state-of-the-art methods on blind video temporal consistency.
Researcher Affiliation Academia Chenyang Lei Yazhou Xing Qifeng Chen The Hong Kong University of Science and Technology
Pseudocode No The paper describes its method using text and mathematical equations, but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes Our source codes are publicly available at github.com/Chenyang LEI/deep-video-prior.
Open Datasets Yes Following the previous work [3, 19], we adopt the DAVIS dataset [29] and the test set collected by Bonneel et al. [3] for evaluation.
Dataset Splits Yes Therefore, we simply select the same epoch (25 or 50 epochs) for all the videos with different length (30 to 200 frames) in a task based on a small validation set up to 5 videos.
Hardware Specification Yes For an 800 480 video frame, our approach costs 80 ms for each iteration during training on Nvidia RTX 2080 Ti.
Software Dependencies No The paper mentions using U-Net architecture, perceptual loss, and the Adam optimizer, but does not specify software dependencies with version numbers (e.g., PyTorch/TensorFlow version, Python version, etc.).
Experiment Setup Yes Implementation details. We use the Adam optimizer [18] and set the learning rate to 0.0001 for all the tasks. The batch size is 1. Dehazing, spatial white balancing, and image enhancement are trained for 25 epochs. Intrinsic decomposition, colorization, style transfer, and Cycle GAN are trained for 50 epochs.