I am trying to train a custom Caffe model for face recognition.
(Refer to Github project here: https://github.com/danduncan/HappyNet)
For that I am using a Docker image for Caffe (which works), in a sharedfolder on my machine:
Docker sharedfolder structure:
execute_3_train_custom_model
datasets/
mean_training_image.binaryproto
training_set_lmdb/
data.mdb
lock.mdb
models/
custom_Model/
deploy.prototxt
EmotiW_VGG_S.caffemodel
loss_history.txt
solver.prototxt
train.prototxt
At training, I'm running a script at command line:
execute_3_train_custom_model:
time ~/caffe/build/tools/caffe train -solver models/Custom_Model/solver.prototxt -weights models/Custom_Model/EmotiW_VGG_S.caffemodel | tee caffe_loss_history.txt
like so:
root@3f3220158436:~/sharedfolder/caffe/docker/image/happyNet# sudo ./execute_3_train_custom_model
Other relevant files:
train.prototxt
name: "CaffeNet"
layers {
name: "training_train"
type: DATA
data_param {
source: "datasets/training_set_lmdb"
backend: LMDB
batch_size: 400
}
transform_param{
mean_file: "datasets/mean_training_image.binaryproto"
}
top: "data"
top: "label"
include {
phase: TRAIN
}
}
layers {
name: "training_test"
type: DATA
data_param {
source: "datasets/validation_set_lmdb"
backend: LMDB
batch_size: 14
}
transform_param{
mean_file: "datasets/mean_training_image.binaryproto"
}
top: "data"
top: "label"
include {
phase: TEST
}
(...)
solver.prototxt
net: "models/Custom_Model/train.prototxt"
# test_iter specifies how many forward passes the test should carry out
test_iter: 1
# Carry out testing every X training iterations
test_interval: 20
# Learning rate and momentum parameters for Adam
base_lr: 0.001
momentum: 0.9
momentum2: 0.999
# Adam takes care of changing the learning rate
lr_policy: "fixed"
# Display every X iterations
display: 10
# The maximum number of iterations
max_iter: 3000
# snapshot intermediate results
snapshot: 100
snapshot_prefix: "snapshot"
# solver mode: CPU or GPU
type: "Adam"
solver_mode: CPU
but when I run the script I get the following traceback:
I0415 01:49:52.260529 148 layer_factory.hpp:77] Creating layer training_test
I0415 01:49:52.260713 148 net.cpp:91] Creating Layer training_test
I0415 01:49:52.260766 148 net.cpp:399] training_test -> data
I0415 01:49:52.260816 148 net.cpp:399] training_test -> label
I0415 01:49:52.260861 148 data_transformer.cpp:25] Loading mean file from: datasets/mean_training_image.binaryproto
F0415 01:49:52.268076 153 db_lmdb.hpp:15] Check failed: mdb_status == 0 (2 vs. 0) No such file or directory
*** Check failure stack trace: ***
@ 0x7f9e61bf7daa (unknown)
@ 0x7f9e61bf7ce4 (unknown)
@ 0x7f9e61bf76e6 (unknown)
@ 0x7f9e61bfa687 (unknown)
@ 0x7f9e6229d0b1 caffe::db::LMDB::Open()
@ 0x7f9e6224d754 caffe::DataReader::Body::InternalThreadEntry()
@ 0x7f9e6072ca4a (unknown)
@ 0x7f9e6050b184 start_thread
@ 0x7f9e60a3137d (unknown)
@ (nil) (unknown)
real 0m1.741s
user 0m0.580s
sys 0m1.230s
I wonder if I'm messing with my path somehow, since I'm using Docker, and I don't know how to debug this.
It seems like you should have two lmdbs:
datasets/training_set_lmdb # which you seem to have
datasets/validation_set_lmdb # where is this one?
When caffe is constructing layer "training_test"
for phase: TEST
you get this error: caffe cannot find datasets/validation_set_lmdb
. Make sure you got it.