整理:AI算法与图像处理
CVPR2022论文和代码整理:https://github.com/DWCTOD/CVPR2022-Papers-with-Code-Demo
ECCV2022论文和代码整理:https://github.com/DWCTOD/ECCV2022-Papers-with-Code-Demo
最新成果demo展示:
南洋理工&南开提出CuDi:曲线蒸馏用于高效可控曝光调整
标题:CuDi: Curve Distillation for Efficient and Controllable Exposure Adjustment
论文:https://arxiv.org/pdf/2207.14273.pdf
代码:https://li-chongyi.github.io/CuDi_files/
ECCV2022 汇总:https://github.com/DWCTOD/ECCV2022-Papers-with-Code-Demo/
摘要:
提出了曲线蒸馏,CuDi,用于有效和可控的曝光调整,而不需要在训练期间配对或不配对的数据。我们的方法继承了有效的低光图像增强方法 Zero-DCE 的零参考学习和基于曲线的框架,进一步加快了推理速度,减小了模型尺寸,并扩展到了可控的曝光调整。改进的推理速度和轻量级模型是通过新颖的曲线蒸馏实现的,该曲线蒸馏通过高阶曲线的切线逼近传统基于曲线的框架中耗时的迭代操作。通过新的自监督空间曝光控制损失实现可控曝光调整,该损失将输出的不同空间区域的曝光水平限制为接近作为输入条件的曝光图的亮度分布。与大多数只能校正曝光不足或曝光过度照片的现有方法不同,我们的方法使用单个模型同时校正曝光不足和曝光过度的照片。值得注意的是,我们的方法可以在输入条件曝光图的指导下全局或局部调整照片的曝光水平,该输入条件曝光图可以在推理阶段预定义或手动设置。通过广泛的实验,我们表明我们的方法因其快速、稳健和灵活的性能而具有吸引力,在真实场景中优于最先进的方法
最新论文整理
ECCV2022
Updated on : 4 Aug 2022
total number : 6
KD-SCFNet: Towards More Accurate and Efficient Salient Object Detection via Knowledge Distillation
- 论文/Paper: http://arxiv.org/pdf/2208.02178
- 代码/Code: https://github.com/zhangjinCV/KD-SCFNet.
Gradient-based Uncertainty for Monocular Depth Estimation
- 论文/Paper: http://arxiv.org/pdf/2208.02005
- 代码/Code: https://github.com/jhornauer/GrUMoDepth.
PolarMOT: How Far Can Geometric Relations Take Us in 3D Multi-Object Tracking?
- 论文/Paper: http://arxiv.org/pdf/2208.01957
- 代码/Code: None
PalQuant: Accelerating High-precision Networks on Low-precision Accelerators
- 论文/Paper: http://arxiv.org/pdf/2208.01944
- 代码/Code: url{https://github.com/huqinghao/PalQuant}.
SuperLine3D: Self-supervised Line Segmentation and Description for LiDAR Point Cloud
- 论文/Paper: http://arxiv.org/pdf/2208.01925
- 代码/Code: https://github.com/zxrzju/SuperLine3D.git.
Learning Prior Feature and Attention Enhanced Image Inpainting
- 论文/Paper: http://arxiv.org/pdf/2208.01837
- 代码/Code: https://github.com/ewrfcas/MAE-FAR
CVPR2022
Updated on : 4 Aug 2022
total number : 3
Negative Frames Matter in Egocentric Visual Query 2D Localization
- 论文/Paper: http://arxiv.org/pdf/2208.01949
- 代码/Code: https://github.com/facebookresearch/vq2d_cvpr
Per-Clip Video Object Segmentation
- 论文/Paper: http://arxiv.org/pdf/2208.01924
- 代码/Code: https://github.com/pkyong95/PCVOS
Combined CNN Transformer Encoder for Enhanced Fine-grained Human Action Recognition
- 论文/Paper: http://arxiv.org/pdf/2208.01897
- 代码/Code: None