Caffe accuracy always 0.5 and loss ~0.69
440 views
Skip to first unread message
Robmosh
Jul 1, 2016, 8:42:46 AM7/1/16
to Caffe Users
Hey there,
i'm training the AlexNet with some little changes to perform a classification and later an object detection on images.
Since i use this net i suffer from the problem that the accuracy stays at 0.5 and the loss at 0.69.
I already googled the problem and have done some of the suggestions, e.g. set bias_filler to 0.5-0.7 instead of 1, shuffle the data again and check wether it is correct or not.
I build the lmdb for training and testing with the convert_imageset from ./build/tools without error with shuffling and resizing.
The net looks like:
layer {
name: "data"
type: "Data"
top: "data"
top: "label"
transform_param {
scale: 0.00392156862745
}
data_param {
source: "/home/rkoch/Bilderkennung/bumper_vs_door/train_lmdb"
batch_size: 200
backend: LMDB
}
}
layer {
name: "conv1"
type: "Convolution"
bottom: "data"
top: "conv1"
convolution_param {
num_output: 32
pad: 2
kernel_size: 5
weight_filler {
type: "gaussian"
std: 0.0001
}
bias_filler {
type: "constant"
value: 0
}
}
}
layer {
name: "relu1"
type: "ReLU"
bottom: "conv1"
top: "conv1"
}
layer {
name: "pool1"
type: "Pooling"
bottom: "conv1"
top: "pool1"
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
layer {
name: "norm1"
type: "LRN"
bottom: "pool1"
top: "norm1"
lrn_param {
local_size: 3
alpha: 0.0001
beta: 0.75
}
}
layer {
name: "conv2"
type: "Convolution"
bottom: "norm1"
top: "conv2"
convolution_param {
num_output: 256
kernel_size: 5
weight_filler {
type: "gaussian"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.7
}
}
}
layer {
name: "relu2"
type: "ReLU"
bottom: "conv2"
top: "conv2"
}
layer {
name: "pool2"
type: "Pooling"
bottom: "conv2"
top: "pool2"
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
layer {
name: "norm2"
type: "LRN"
bottom: "pool2"
top: "norm2"
lrn_param {
local_size: 5
alpha: 0.0001
beta: 0.75
}
}
layer {
name: "conv3"
type: "Convolution"
bottom: "norm2"
top: "conv3"
convolution_param {
num_output: 384
pad: 1
kernel_size: 3
weight_filler {
type: "gaussian"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.5
}
}
}
layer {
name: "relu3"
type: "ReLU"
bottom: "conv3"
top: "conv3"
}
layer {
name: "conv4"
type: "Convolution"
bottom: "conv3"
top: "conv4"
convolution_param {
num_output: 384
pad: 1
kernel_size: 3
group: 2
weight_filler {
type: "gaussian"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.5
}
}
}
layer {
name: "relu4"
type: "ReLU"
bottom: "conv4"
top: "conv4"
}
layer {
name: "conv5"
type: "Convolution"
bottom: "conv4"
top: "conv5"
convolution_param {
num_output: 256
pad: 1
kernel_size: 3
weight_filler {
type: "gaussian"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.5
}
}
}
layer {
name: "relu5"
type: "ReLU"
bottom: "conv5"
top: "conv5"
}
layer {
name: "pool5"
type: "Pooling"
bottom: "conv5"
top: "pool5"
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
layer {
name: "full6"
type: "InnerProduct"
bottom: "pool5"
top: "full6"
inner_product_param {
num_output: 4096
weight_filler {
type: "gaussian"
std: 0.001
}
bias_filler {
type: "constant"
value: 1
}
}
}
layer {
name: "relu6"
type: "ReLU"
bottom: "full6"
top: "full6"
}
layer {
name: "full8"
type: "InnerProduct"
bottom: "full6"
top: "full8"
inner_product_param {
num_output: 1000
weight_filler {
type: "gaussian"
std: 0.01
}
bias_filler {
type: "constant"
value: 0.5
}
}
}
layer {
name: "score"
type: "InnerProduct"
bottom: "full8"
top: "score"
inner_product_param {
num_output: 2
weight_filler {
type: "xavier"
}
}
}
layer {
name: "loss"
type: "SoftmaxWithLoss"
bottom: "score"
bottom: "label"
top: "loss"
}
Can somebody help me with this problem?
Thanks,
Robert
Shahmi
Feb 27, 2017, 11:19:04 AM2/27/17
to Caffe Users
Did you resolve the problem anyway? I'm stuck with the similar problem and curious how you went about it.
Thank you.
Thank you.
Message has been deleted
sayad yaghoobi
Apr 27, 2017, 9:36:15 AM4/27/17
to Caffe Users
i'm stuck in the same issue too. i checked two network, both alexnet and nin. but both of them give me the same loss and accuracy.
if anyone has solved this issue, please confirm me to. i'll be grateful
Guofeng Hu
May 3, 2017, 9:51:55 PM5/3/17
to Caffe Users
Hi，
My model's accuracy is something like you. My train acccuracy is about 0.90, but the test accuracy is about 0.55. When I increase the dropout from 0.5 to 0.6, it does little work. I don't know why this happens.
在 2016年7月1日星期五 UTC+8下午8:42:46，Robmosh写道：
在 2016年7月1日星期五 UTC+8下午8:42:46，Robmosh写道：
Reply all
Reply to author
Forward
0 new messages