Caffe
visionfans
研究方向:计算机视觉,模式识别,机器学习
展开
-
fully connected layer as 1x1 convolution
Quoting Lecun’s post: In Convolutional Nets, there is no such thing as “fully-connected layers”. There are only convolution layers with 1x1 convolution kernels and a full connection table. It’s a原创 2015-09-07 19:07:53 · 2045 阅读 · 0 评论 -
Caffe blob reshape
Release original memory and hold new: template <typename Dtype> void Blob<Dtype>::Reshape(const vector<int>& shape) { CHECK_LE(shape.size(), kMaxBlobAxes); count_ = 1; shape_.resize(shape.size())原创 2015-09-12 21:54:11 · 5282 阅读 · 0 评论 -
Notes on Caffe layers
to finish.References Developing new layers Switch LayerType to string for extensibility #1685 Layer type is a string #1694原创 2015-09-12 13:32:33 · 757 阅读 · 0 评论 -
caffe layer `EltwiseLayer`
coeff should be empty or be the same size with bottom coeff only be used with SUM operation top and all bottoms are the same size operations top[0] = PROD(bottom[:]) top[0] = SUM(coeff[i]*bottom[i])原创 2015-11-04 00:26:04 · 2720 阅读 · 0 评论 -
caffe training tricks
Implement the model that won the classification task of ImageNet 2013 #33 choosing batch sizes and tuning sgd #218 How to train imagenet with reduced memory and batch size? #430 Originally base_lr = 0原创 2015-11-01 20:39:08 · 904 阅读 · 0 评论 -
Caffe cuDNN max-pooling with in-place dropout
dropout in place incompatible with max pooling #117 Fix Max pooling layer to use a mask #162 Finish up max pooling with a mask from @sguada #448 In-place computation can break gradient computation #201原创 2015-11-06 22:24:35 · 898 阅读 · 0 评论 -
caffe layer
看Caffe/wiki/Development,同时对照ArgMaxLayer来看! layer的声明要写到: l common_layers.hpp l data_layers.hpp l loss_layers.hpp l neuron_layers.hpp l vision_layers.hpp 要实现以下函数: l LayerSetUp l Re原创 2015-10-26 23:34:02 · 1240 阅读 · 0 评论 -
caffe loss NaN
by Yangqing on 14 May 2014 For a sanity check, try running with a learning rate 0 to see if any nan errors pop up (they shouldn’t, since no learning takes place). If data is not initialized well, it原创 2015-11-01 20:28:55 · 2572 阅读 · 0 评论