计算机视觉

• 转换和增强图像
  • 支持的输入类型和约定
  • 数据类型和预期值范围
  • 性能考虑
  • Torchscript 支持
• 模型和预训练权重
  • 关于预训练权重的一般信息
  • 初始化预训练模型
  • 使用预训练模型
  • 列出和检索可用的模型
  • 使用来自 Hub 的模型
• 快速上手 torchvision.transforms.v2
  • v2 基础
  • 简单的分类管道
  • 检测、分割、视频
  • TVTensors 是什么？
  • 需要传递什么作为输入？
  • 转换和数据集的兼容性
    • 内置的数据集
• Illustration of transforms
  • Geometric Transforms
    • Pad
    • Resize
    • CenterCrop
    • FiveCrop
    • RandomPerspective
    • RandomRotation
    • RandomAffine
    • ElasticTransform
    • RandomCrop
    • RandomResizedCrop
  • Photometric Transforms
    • Grayscale
    • ColorJitter
    • GaussianBlur
    • RandomInvert
    • RandomPosterize
    • RandomSolarize
    • RandomAdjustSharpness
    • RandomAutocontrast
    • RandomEqualize
  • Augmentation Transforms
    • AutoAugment
    • RandAugment
    • TrivialAugmentWide
    • AugMix
  • Randomly-applied Transforms
    • RandomHorizontalFlip
    • RandomVerticalFlip
    • RandomApply
• Transforms v2: End-to-end object detection/segmentation example
  • Dataset preparation
  • Transforms
  • Data loading and training loop
  • Training References
• How to use CutMix and MixUp
  • Pre-processing pipeline
  • Where to use MixUp and CutMix
    • After the DataLoader
    • As part of the collation function
  • Non-standard input format
• How to write your own v2 transforms
  • Just create a nn.Module and override the forward method
  • Supporting arbitrary input structures
• TVTensors FAQ
  • What are TVTensors?
  • What can I do with a TVTensor?
  • How do I construct a TVTensor?
    • Using the constructor
    • Using tv_tensors.wrap()
  • I had a TVTensor but now I have a Tensor. Help!
    • But I want a TVTensor back!
    • Why is this happening?
    • Exceptions
• How to write your own TVTensor class
  • Parameter forwarding, and ensuring future compatibility of your kernels
• Optical Flow: Predicting movement with the RAFT model
  • Reading Videos Using Torchvision
  • Estimating Optical flow using RAFT
  • Visualizing predicted flows
  • Bonus: Creating GIFs of predicted flows
• Repurposing masks into bounding boxes
  • Masks
  • Converting Masks to Bounding Boxes
  • Converting Segmentation Dataset to Detection Dataset
• Torchscript support
• Video API
  • 1. Introduction: building a new video object and examining the properties
  • 2. Building a sample read_video function
  • 3. Building an example randomly sampled dataset (can be applied to training dataset of kinetics400)
  • 4. Data Visualization
• Visualization utilities
  • Visualizing a grid of images
  • Visualizing bounding boxes
  • Visualizing segmentation masks
    • Semantic segmentation models
    • Instance segmentation models
  • Visualizing keypoints
• feature_extraction