Video Super-Resolution Quantization (Time:2023.07.07-2023.08.07)
Paper Reading
Dynamic Network Quantization for Efficient Video Inference (ICCV2021)
Feat: selects optimal precision for each frame conditioned on the input for efficient video recognition
ResQ: Residual Quantization for Video Perception (ICCV2023)
Feat: difference in network activations between two neighboring frames, exhibit properties that make them highly quantizable
QuantSR: Accurate Low-bit Quantization for Efficient Image Super-Resolution (NIPS2023)
To overcome the representation homogeneity caused by quantization in the network, we introduce the Redistribution-driven Learnable Quantizer (RLQ). This is accomplished through an inference-agnostic efficient redistribution design, which adds additional information in both forward and backward passes to improve the representation ability of quantized networks. (为了克服网络中量化造成的表示同质性，我们引入了重分布驱动的可学习量化器 (RLQ)。这是通过与推理无关的高效重分布设计实现的，它在前向和后向传递中添加了额外信息，以提高量化网络的表示能力。)
Furthermore, to achieve flexible inference and break the upper limit of accuracy, we propose the Depth-dynamic Quantized Architecture (DQA). Our DQA allows for the trade-off between efficiency and accuracy during inference through weight sharing.(此外，为了实现灵活的推理并突破准确率的上限，我们提出了深度动态量化架构（DQA）。我们的DQA通过权重共享，实现了推理过程中效率和准确率之间的平衡。)
Knowledge Distillation for Optical Flow-Based Video Superresolution (JCSE2023)
Feat: Video super-resolution; Optical flow; Knowledge distillation;
EDVR: Video Restoration with Enhanced Deformable Convolutional Networks (NTIRE2019)
leverage temporal redundancies to accelerate video processing
Towards High Performance Video Object Detection for Mobiles (MSRA_arxiv2018)
Temporally Distributed Networks for Fast Video Semantic Segmentation (CVPR2020)