CNN

Convolutional Neural Networks (CNNs), inspired by the structure of the visual cortex, extract local features by sliding convolutional kernels across images, with the parameter sharing mechanism significantly reducing computational load.

Image recognition in agricultural scenarios is primarily based on convolutional neural networks (CNN) in deep learning and their variants, such as model architectures including ResNet, EfficientNet, and Vision Transformer.

Vision Transformer (ViT) emerged in 2020, marking the entry of Transformers into the field of computer vision.

In the deep learning era, end-to-end models based on CNN and RNN (such as the CRNN architecture) significantly improved OCR recognition capabilities, and in recent years the Transformer architecture has also been introduced into the OCR field.

2015年Gatys等人提出神经风格迁移，通过卷积神经网络的Gram矩阵来匹配内容图与风格图的特征分布

The core design principles of CNNs are local receptive fields and weight sharing, which make their parameter count far smaller than fully connected networks and give them inherent translation invariance.

ViT (Vision Transformer), proposed by Google in 2020, demonstrated that after splitting images into 16×16 patches, the Transformer can match or even exceed CNN performance on image classification tasks.

宠物品种识别可使用卷积神经网络（CNN）模型如ResNet、EfficientNet实现

特征提取阶段模型通过卷积神经网络（CNN）或视觉Transformer（ViT）等架构将原始像素转化为抽象特征向量

CNN能从原始像素中自动提取层次化特征，Transformer能从原始文本序列中学习语义表示，大幅降低了对手工特征工程的依赖

ViT与卷积神经网络不同，从第一层起就能捕获图像中任意两个位置之间的关联

CNNs and their subsequent variants (such as ResNet, YOLO, and U-Net) form the skeleton of modern computer vision.

In image classification, a CNN usually ends with a fully connected layer that outputs class probabilities; in object detection, the network predicts bounding box offsets and classes on the feature map; in image segmentation, an encoder-decoder structure is often used to upsample low-resolution semantic features back to the original image size to achieve pixel-level prediction.

CNN (convolutional neural network) is the key model architecture that makes the three types of tasks—image classification, object detection, and image segmentation—practically feasible, and its core idea is to use learnable convolution kernels to slide and extract local features, then stack them layer by layer to combine them into high-level semantics.

After Hugging Face released BERT and CNN-based models, natural language classification capability became further democratized; the open-sourcing of the Transformers library, TensorFlow, and PyTorch allowed small teams or individuals to quickly obtain decent results with the right data and a moderate level of fine-tuning knowledge.

阿弗莱克指出CNN是Transformer的'前身'，区别在于Transformer能同时进行更多的计算。

把CNN称为Transformer的'前身'是一种通俗但不精确的说法，两者是不同时期主导视觉任务的架构范式，并非取代关系。

在影视特效中，CNN常被用于自动抠像（matting）、运动跟踪、超分辨率重建等任务；机器学习算法可在半透明物体、毛发边缘、运动模糊等困难区域自动估计前景和背景的混合比例。

The typical pipeline of font embedding includes: data preparation (collecting TTF/OTF font files and rendering characters), feature extraction (encoding into vectors with a CNN or VAE), training the embedding space (contrastive learning or autoencoder reconstruction loss), dimensionality reduction visualization (t-SNE, UMAP or PCA), and observing the structure.

Warner Bros. Discovery owns content assets and brands including Warner Bros. Pictures, HBO, CNN, and Discovery.