315386775/cv_goodidea

We're IDEA Public StoreHouse. Please keep the ideas coming.

Introduction

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.

Papers

Technical blog

• [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]

Survey

• 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

AI视觉-实际应用场景下必解决的点

图像低质量/模糊/压缩等问题

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

• AdaFace: Quality Adaptive Margin for Face Recognition (CVPR2022 Oral) [paper] [code]

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。

backbone

git clone git
pip3 install -v -e .  # or  python3 setup.py develop

Deployment

1. MegEngine in C++ and Python
2. ONNX export and an ONNXRuntime
3. TensorRT in C++ and Python
4. ncnn in C++ and Java
5. OpenVINO in C++ and Python

AI-视觉实际场景下

Cite CV_GoodIdea

If you use CV_GoodIdea in your research, please cite our work by using the following:

 @article{,
  title={},
  author={},
  journal={},
  year={2022}
}