Train and Test LeNet on your own dataset

the purpose of this topic is：

1. how to train LeNet (defined in examples/mnist/lenet.prototxt) using your own dataset( jpg images of the same size or preprocessed ) 
2. test using the trained LeNet.

step 1: convert the image files to lmdb database format

suppose we have prepared all the images in .jpg format under some folders; train.txt and val.txt that contains the [filename label] of the training and testing data (label: 0~#of labels-1)

reference  examples/imagenet/create_imagenet.sh

change the following path:

• EXAMPLE = /path/to/subfolder/under/examples/to/store/lmdb file
• DATA = /path/to/folder/containing/train.txt & val.txt
• TRAIN_DATA_ROOT=/path/to/folder/omitted/in/train.txt
• TRAIN_DATA_ROOT=/path/to/folder/omitted/in/val.txt
• GLOG….$EXAMPLE/…_train(val)_lmdb (rename the database for training and testing)
• if the input images are gray, add ” – -gray \” after ” – -shuffle \”

step 2: modify the lenet model to adapt the input file and filepath

reference examples/mnist/lenet.prototxt, lenet_train_test.prototxt, lenet_solver.prototxt, train_lenet.sh

• lenet.prototxt and lenet_train_test.prototxt:

1.       input_dim: (batchsize, #channels, imagesize1,imagesize2)
2.       inner_product_param: num_output: #labels
3.       “data” layer: phase: TRAIN/TEST  source: “…”(path to training/testing lmdb database)

! the default path in “data” layer omit caffe_root, and you need to call ./build/tools/caffe at caffe_root !

• lenet_solver.prototxt:

1. test_iter:  #test batches ( the total test images evaluated in TRAIN phase = test_iter*TEST batch size )
2. base_lr: initial learning rate ( default:.01, change to a smaller number if getting NAN loss in training )
3. snapshot_prefix: “…”(filepath to snapshot saved)
4. snapshot: # of iterations per snapshot
5. max_iter:  # of iterations

• train_lenet.sh:

1. –solver = *_solver.prototxt

step 3: train the model

run ./examples/…./train_*.sh in the terminal

step 4: test the model

cd to caffe_root, run:

./build/tools/caffe test -model /path/to/*_train_test.prototxt -weights/path/to/trained/caffe/caffemodel -iterations #test images/batch_size

---

use new dataset for modified network

• clear lmdb database
• rename old files: data/*/train.txt val.txt  examples/*/*.caffemodel *.solverstate
• update the parameters in examples/*/train_*net.sh if necessary
• create new train.txt val.txt
• run examples/*/create_*.sh, train_*.sh