Image classification

July 6, 2018 ยท View on GitHub

Description

Assigning a single class label to the whole input image.

Metrics:

  • Top-1 Accuracy
  • Top-5 Accuracy

Results

ImageNet

Single model, single crop. Top-5 results by Acc@5 are highlighted in boldface as well.

Dataset description

ModelAcc@1Acc@5WeightsPaperCode
DenseNet-12182.85892.290PyTorchpaperPyTorch
DenseNet-16182.73693.798PyTorchpaperPyTorch
DenseNet-16982.69393.150PyTorchpaperPyTorch
DenseNet-20182.56693.660PyTorchpaperPyTorch
Inception-v181.32089.600TensorFlowpaperTensorFlow
Inception-v281.30491.800TensorFlowpaperTensorFlow
Inception-v380.40093.900TensorFlow PyTorchpaperTensorFlow PyTorch
Inception-v480.20095.200TensorFlowpaperTensorFlow
Inception-v480.17094.930PyTorchpaperPyTorch
Inception-ResNet-v280.06095.300TensorFlowpaperTensorFlow
Inception-ResNet-v278.88895.234PyTorchpaperPyTorch
NASNet-A-Large78.42896.163TensorFlowpaperTensorFlow
NASNet-A-Large78.00096.086PyTorchpaperPyTorch
PNASNet-5-Large77.56096.182TensorFlowpaperTensorFlow
PNASNet-5-Large77.43895.992PyTorchpaperPyTorch
SENet15477.42095.530CaffepaperCaffe
FBResNet-15277.38693.594PyTorchpaperPyTorch
SENet15476.20095.498PyTorchpaperPyTorch
Xception74.98094.292PyTorchpaperPyTorch
ResNet-10173.90093.672PyTorchpaperPyTorch

PASCAL VOC Classification

TODO: add dataset description

Dataset description

ModelScoreWeightsPaperCode
Model name0.5link to the weights of a trained modelpaperlink to implementation

Resources

TitleTypeDescription
pretrained-models.pytorchRepositoryRepo dedicated to reproduction of research paper results and collect pretrained convolutional nets for classification (PyTorch).
tf/research/slimRepositoryConvolutional nets for classification pretrained on ImageNet (TF Slim)