Home

Awesome

Densely Connected Convolutional Network (DenseNet)

paper: Densely Connected Convolutional Networks

This is the MXNet implement of DenseNet with pretrained model, therefor you can fine-tune in the pretrained model for your own dataset.

Do as follows:

NetworkTop-1 errorMXNet model
DenseNet-121 (k=32)25.16Google Drive (32.3MB)
DenseNet-169 (k=32)23.74Google Drive (57.3MB)
DenseNet-201 (k=32)22.54Google Drive (81.0MB)
DenseNet-161 (k=48)22.28Google Drive (115.7MB)

These pretrained models are manually converted from https://github.com/shicai/DenseNet-Caffe ,put the pretrained model under /DenseNet/model/ file.

If you want to use .rec file to train your model:

sh run_train_rec.sh

If you want to use .lst file and image to train your model:

sh run_train_lst.sh