All Projects → revilokeb → vgg16_batchnorm

revilokeb / vgg16_batchnorm

Licence: other
VGG16 architecture with BatchNorm

Projects that are alternatives of or similar to vgg16 batchnorm

Image-Classification
Pre-trained VGG-Net Model for image classification using tensorflow
Stars: ✭ 29 (+107.14%)
Mutual labels:  vgg16, pre-trained
Resnet Cifar10 Caffe
ResNet-20/32/44/56/110 on CIFAR-10 with Caffe
Stars: ✭ 161 (+1050%)
Mutual labels:  caffe, vgg16
Cnn Models
ImageNet pre-trained models with batch normalization for the Caffe framework
Stars: ✭ 355 (+2435.71%)
Mutual labels:  caffe, vgg16
Mtcnn caffe
Simple implementation of kpzhang93's paper from Matlab to c++, and don't change models.
Stars: ✭ 244 (+1642.86%)
Mutual labels:  caffe
Senet
Squeeze-and-Excitation Networks
Stars: ✭ 2,850 (+20257.14%)
Mutual labels:  caffe
caffe-conf-matrix
Python layer for the Caffe deep learning framework to compute the accuracy and the confusion matrix.
Stars: ✭ 20 (+42.86%)
Mutual labels:  caffe
Image2LMDB
Convert image folder to lmdb, adapted from Efficient-PyTorch
Stars: ✭ 58 (+314.29%)
Mutual labels:  caffe
Colorization
Automatic colorization using deep neural networks. "Colorful Image Colorization." In ECCV, 2016.
Stars: ✭ 2,791 (+19835.71%)
Mutual labels:  caffe
MachineLearning
An easy neural network for Java!
Stars: ✭ 125 (+792.86%)
Mutual labels:  weights
List-CoreML-Models
A Big Awesome List CoreML Models.
Stars: ✭ 120 (+757.14%)
Mutual labels:  caffe
image-segmentation
Mask R-CNN, FPN, LinkNet, PSPNet and UNet with multiple backbone architectures support readily available
Stars: ✭ 62 (+342.86%)
Mutual labels:  pre-trained
RoundWidget
Java 实现的 Android 原生圆角矩形控件,圆角 FrameLayout 和圆角 ImageView
Stars: ✭ 86 (+514.29%)
Mutual labels:  weights
Cnn face detection
Implementation based on the paper Li et al., “A Convolutional Neural Network Cascade for Face Detection, ” 2015 CVPR
Stars: ✭ 251 (+1692.86%)
Mutual labels:  caffe
mbus
基于RabbitMQ简单实现验证码识别平台,训练网络模型智能识别图形验证码
Stars: ✭ 37 (+164.29%)
Mutual labels:  caffe
Cnnforandroid
The Convolutional Neural Network(CNN) for Android
Stars: ✭ 245 (+1650%)
Mutual labels:  caffe
keras-flask-deploy-webapp
😺 Pretty & simple image classifier app template. Deploy your own trained model or pre-trained model (VGG, ResNet, Densenet) to a web app using Flask in 10 minutes.
Stars: ✭ 1,078 (+7600%)
Mutual labels:  pre-trained
Retinaface
Reimplement RetinaFace use C++ and TensorRT
Stars: ✭ 241 (+1621.43%)
Mutual labels:  caffe
Cgnl Network.pytorch
Compact Generalized Non-local Network (NIPS 2018)
Stars: ✭ 252 (+1700%)
Mutual labels:  caffe
tf-semantic-segmentation-FCN-VGG16
Semantic segmentation for classifying road. "Fully Convolutional Networks for Semantic Segmentation (2015)" implemented using TF
Stars: ✭ 30 (+114.29%)
Mutual labels:  vgg16
triplet
Re-implementation of tripletloss function in FaceNet
Stars: ✭ 94 (+571.43%)
Mutual labels:  caffe

vgg16_batchnorm in Caffe

Caffe weights of VGG16 architecture with BatchNorm

Here I have trained the VGG16 architecture with interleaved batch-norm layers on the ILSVRC12 dataset in Caffe. I have started the training with all available weights initialized from the original VGG16 contribution (http://www.robots.ox.ac.uk/%7Evgg/research/very_deep/). I am using subtraction of channel mean as in the original publication and also keep biases (although they are not necessary).

The training dataset is the full ILSVRC12 training data resized to shorter side 256px. During training I do mirroring and random cropping (224 x 224) but no other data augmentation, see also train_val_vgg16bn.prototxt.

After 28 epochs the single center crop (224 x 224) validation accuracy is as follows (caffe test -model train_val_vgg16bn.prototxt -weights vgg16bn_2_iter_1120000.caffemodel -gpu 0 -iterations 6250):

  • Top 1: 70.44%
  • Top 5: 89.94%

This compares with the following single center crop (224 x 224) validation accuracy of the orginal weights from http://www.robots.ox.ac.uk/%7Evgg/research/very_deep/:

  • Top 1: 70.38%
  • Top 5: 89.38%

Link to the weights: https://drive.google.com/drive/folders/0B1qLpHDbczM2NGJwMHFDLTdNZkU?usp=sharing

Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].