Skip to content

Latest commit

 

History

History
68 lines (47 loc) · 1.52 KB

RELEASE.md

File metadata and controls

68 lines (47 loc) · 1.52 KB
Raw

Release Note

MindSpore Computer Vision 0.0.1

Models

mindcv.models now expose num_classes and in_channels as constructor arguments:

  • Add DenseNet models and pre-trained weights
  • Add GoogleNet models and pre-trained weights
  • Add Inception V3 models and pre-trained weights
  • Add Inception V4 models and pre-trained weights
  • Add MnasNet models and pre-trained weights
  • Add MobileNet V1 models and pre-trained weights
  • Add MobileNet V2 models and pre-trained weights
  • Add MobileNet V3 models and pre-trained weights
  • Add ResNet models and pre-trained weights
  • Add ShuffleNet V1 models and pre-trained weights
  • Add ShuffleNet V2 models and pre-trained weights
  • Add SqueezeNet models and pre-trained weights
  • Add VGG models and pre-trained weights
  • Add ViT models and pre-trained weights

Dataset

mindcv.data now expose:

  • Add Mnist dataset
  • Add FashionMnist dataset
  • Add Imagenet dataset
  • Add CIFAR10 dataset
  • Add CIFAR100 dataset

Loss

mindcv.loss now expose:

  • Add BCELoss
  • Add CrossEntropyLoss

Optimizer

mindcv.optim now expose:

  • Add SGD optimizer
  • Add Momentum optimizer
  • Add Adam optimizer
  • Add AdamWeightDecay optimizer
  • Add RMSProp optimizer
  • Add Adagrad optimizer
  • Add Lamb optimizer

Learning_Rate Scheduler

mindcv.scheduler now expose:

  • Add WarmupCosineDecay learning rate scheduler
  • Add ExponentialDecayLR learning rate scheduler
  • Add Constant learning rate scheduler

Release

mindcv-0.0.1.apk

mindcv-0.0.1-py3-none-any.whl.sha256

mindcv-0.0.1-py3-none-any.whl