Skip to first unread message
Praveen O
Dec 27, 2015, 2:00:06 PM12/27/15
to Caffe Users
The training set consists of images (3x256x256), along with a mask label (256x256 - can be viewed as a grayscale image). I created LMDB for both the images and masks using create_imagenet.sh
Is my network definition correct:
train.prototxt
layer {
name: "data"
type: "Data"
top: "data"
transform_param {
scale: 0.00392156862745
mirror: true
}
data_param {
source: "images/train_lmdb"
batch_size: 8
backend: LMDB
}
}
layer {
name: "mask"
type: "Data"
top: "mask"
transform_param {
scale: 0.00392156862745
mirror: true
}
data_param {
source: "images/mask_point1_lmdb"
batch_size: 8
backend: LMDB
}
}
layer {
name: "conv"
type: "Convolution"
bottom: "data"
top: "conv"
convolution_param {
num_output: 16
kernel_size: 3
weight_filler {
type: "gaussian"
std: 0.1
}
}
}
layer {
name: "conv_bn"
type: "BatchNorm"
bottom: "conv"
top: "conv_bn"
}
layer {
name: "relu1"
type: "ReLU"
bottom: "conv_bn"
top: "conv_bn"
}
layer {
name: "deconv"
type: "Deconvolution"
bottom: "conv_bn"
top: "deconv"
convolution_param {
num_output: 1
kernel_size: 3
weight_filler {
type: "gaussian"
std: 0.1
}
}
}
layer {
name: "deconv_bn"
type: "BatchNorm"
bottom: "deconv"
top: "deconv_bn"
}
layer {
name: "logistic"
type: "Sigmoid"
bottom: "deconv_bn"
top: "deconv_bn"
}
layer {
name: "loss"
type: "EuclideanLoss"
bottom: "mask"
bottom: "deconv_bn"
top: "loss"
}
solver.prototxt
# The train/test net protocol buffer definition
net: "train/point1.prototxt"
test_state: { stage: 'test-on-train' }
test_iter: 600
test_interval: 600
test_compute_loss: true
type: "AdaDelta"
gamma: 0.0001
base_lr: 0.1
momentum: 0.95
delta: 1e-6
weight_decay: 0.0005
power: 0.75
lr_policy: "fixed"
display: 600
max_iter: 24000
snapshot_prefix: "model_1"
solver_mode: GPU
I see this error :
I1227 10:44:38.112040 2063626240 layer_factory.hpp:77] Creating layer data
I1227 10:44:38.112077 2063626240 net.cpp:106] Creating Layer data
I1227 10:44:38.112082 2063626240 net.cpp:411] data -> data
F1227 10:44:38.115083 528384 data_reader.cpp:98] Check failed: new_queue_pairs_.size() == 0 (1 vs. 0)
Etienne Perot
Feb 11, 2016, 12:09:22 PM2/11/16
to Caffe Users
I have the exact same error...when using newest caffe... did you solve your problem?
Message has been deleted
Etienne Perot
Feb 11, 2016, 12:27:04 PM2/11/16
to Caffe Users
EDIT : i have solved this by putting the validation data layers for labels AND image.
layer {
name: "image"
type: "Data"
top: "Data1"
include{phase: TRAIN}
data_param {
source: "/home/valeodevbox/SegNet/mscoco_40k/train/image.lmdb"
batch_size: 32
backend: LMDB
}
}
layer {
name: "image"
type: "Data"
top: "Data1"
include{phase: TEST}
data_param {
source: "/home/valeodevbox/SegNet/mscoco_40k/test/image.lmdb"
batch_size: 32
backend: LMDB
}
layer {
name: "label"
type: "Data"
top: "Data2"
include{phase: TRAIN}
data_param {
source: "/home/valeodevbox/SegNet/mscoco_40k/train/label.lmdb"
batch_size: 32
backend: LMDB
}
}
layer {
name: "label"
type: "Data"
top: "Data2"
include{phase: TEST}
data_param {
source: "/home/valeodevbox/SegNet/mscoco_40k/test/label.lmdb"
batch_size: 32
backend: LMDB
}
}
Reply all
Reply to author
Forward
0 new messages