ECCV18_Lifelong_Learning

Thi repository is for the paper "Lifelong Learning via Progressive Distillation and Retrospection".

Instructions

1. Install the dependencies for Caffe according to the official instructions and modify the ./lll-caffe/Makefile.config.
2. Compile the ./lll-caffe and make a soft link to the ./lll-data.

cd lll-caffe
make -j  12
ln -s ../lll-data data

3. Download the pretrained models from the following links and put them in the ./lll-models.

Link1 (Password: wcdfwu)

4. Generate the lmdb for each dataset in the ./lll-data according to the provided image list.

#Take Flowers as an example
./lll-data/data/flowers_train_lmdb #train set
./lll-data/data/flowers_test_lmdb #test set
./lll-data/data/flowers_seed1_uniform5_train_lmdb #subset from train set for Retrospection

5. ImageNet2Flowers is taken as an example to illustrate the usage of the code. Please refer to imagenet2flowers.sh for the details, including training and evaluation.