Pre-trained model list

ImageNet: a large corpus of 1 million natural images, divided into 1000 categories.

1000 categories: synset.txt chinese synset.txt

ResNet: Deep Residual Learning for Image Recognition

Description: Input value of grayscale is ranged 0 to 1 without any pre-processing.

  • resnet-18 params resnet-18 symbol
  • resnet-34 params resnet-34 symbol
  • resnet-50 params resnet-50 symbol
  • resnet-101 params resnet-101 symbol
  • resnet-152 params resnet-152 symbol
  • resnet-200 params resnet-200 symbol

    ResNext: Aggregated Residual Transformations for Deep Neural Networks

    Description: Input value of grayscale is ranged 0 to 1 without any pre-processing.

  • resnext-101-64x4d params resnext-101-64x4d symbol

    Inception net: Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift

    Description: Input value of grayscale is ranged 0 to 255 with a subtracted value of [123.68 116.78 103.94].

  • Inception-BN params Inception-BN symbol

    Mobile net (v2): MobileNetV2: Inverted Residuals and Linear Bottlenecks

    Description: Input value of grayscale is ranged 0 to 255 with a subtracted value of [123.68 116.78 103.94].

  • mobilev2 params mobilev2 symbol