We're IDEA Public StoreHouse. Please keep the ideas coming.
As I continue to absorb as much knowledge as I can, I look forward to each new idea that pops into my head.
I love the process of stumbling upon an idea, hashing it out.
It really is an exciting process. You will gain a wealth of knowledge.
- [English Blog] Transformers in Vision [Link]
- [Chinese Blog] 3W字长文带你轻松入门视觉transformer [Link]
- [Chinese Blog] Vision Transformer 超详细解读 (原理分析+代码解读) [Link]
- A Survey of Visual Transformers [paper] - 2021.11.30
- Transformers in Vision: A Survey [paper] - 2021.02.22
- A Survey on Visual Transformer [paper] - 2021.1.30
- A Survey of Transformers [paper] - 2020.6.09
图像质量是一个属性的组合,表明一个图像如何如实地捕获原始场景。影响图像质量的因素包括亮度、对比度、锐度、噪声、色彩一致性、分辨率、色调再现等。
低质量图像给AI在实际场景下的任务带来极大的挑战,究其根本是其会降低目标的可辨别性。
LOSS_IDEA
出发点: 从度量学习的角度考虑提升特征表示的鉴别性. 新思路: 简单或困难样本的相对重要性应该基于样本的图像质量来给定。 解决方案:通过用feature norms来近似图像质量,在提升低质量图像的识别精度的同时,也没有损失高质量图像的精度。 实验结论:通过对9个不同质量的数据集(LFW、CFP-FP、CPLFW、AgeDB、CALFW、IJB-B、IJB-C、IJB-S和TinyFace)的广泛评估,验证了该方法的有效性。
git clone git
pip3 install -v -e . # or python3 setup.py develop
- MegEngine in C++ and Python
- ONNX export and an ONNXRuntime
- TensorRT in C++ and Python
- ncnn in C++ and Java
- OpenVINO in C++ and Python
AI-视觉实际场景下
If you use CV_GoodIdea in your research, please cite our work by using the following:
@article{,
title={},
author={},
journal={},
year={2022}
}