Follow the instructions in run_me.sh, which details every step in the pipeline.
Assume the testing data are placed under ./data/Test_images_distribute/, the training and validation CENTRIST feature are under ./data/Train_Val_CENTRIST_Feat/ and the testing CENTRIST feature are under ./data/Test_CENTRIST_blocks_4_distribute/
To get the list of testing images ./data/list_all_test.txt:
gunzip ./data/list_all_test.txt.gz
Any face detector is OK. To reproduce the result, use the detection results at ./data/det_rects_test.txt:
gunzip ./data/det_rects_test.txt.gz
cd scripts; matlab -nodesktop -nosplash -r "crop_det_test; exit"; cd -;
Any face landmark detector is OK. To reproduce the result, use the detection results at ./data/landmark_pts.gz:
tar -xvzf ./data/landmark_pts.gz -C ./data
cp ./data/landmark_pts/* ./data/det_crop_test/ -v
cd scripts; matlab -nodesktop -nosplash -r "align_faces; exit"; cd -;
cd scripts; matlab -nodesktop -nosplash -r "generate_list; exit"; cd -;
cd scripts; matlab -nodesktop -nosplash -r "process_scene_feat_testing; exit"; cd -;
git clone https://github.com/BVLC/caffe.git
cp scripts/feat_extract_to_file.cpp ./caffe/tools/
cd caffe; mkdir build; cd build; cmake ../; make -j; cd ../../; # make sure caffe builds well
./caffe/build/tools/feat_extract_to_file ./models/model1.caffemodel ./models/mem_test.prototxt pool5 \
data/list_det_crop_align_filled.txt ./data/ ./data/feat1.txt
./caffe/build/tools/feat_extract_to_file ./models/model2.caffemodel ./models/mem_test.prototxt pool5 \
data/list_det_crop_align_filled.txt ./data/ ./data/feat2.txt
Install apollocaffe, instructions available at http://apollocaffe.com/:
git clone http://github.com/Russell91/apollocaffe.git # make sure apollocaffe builds well
And run
python scripts/test_lstm.py --config models/config.json --gpu 0
python scripts/test_lstm_ord.py --config models/config.json --gpu 0