./download_dataset.sh
./train_fcn32s.py -g 0
./train_fcn16s.py -g 0
./train_fcn8s.py -g 0
./train_fcn8s_atonce.py -g 0
wget https://raw.githubusercontent.com/wkentaro/dotfiles/1.2.1/local/bin/view_log
chmod u+x view_log
./view_log logs/XXX/log.csv
PyTorch implementation is faster for static inputs and slower for dynamic ones than Chainer one at test time.
(In the previous performance, Chainer one was much slower, but it was fixed via wkentaro/fcn#90.)
# Titan X (Pascal)
# chainer==2.0.2
# pytorch==0.2.0.post2
# pytorch-fcn==1.7.0
% cd examples/voc
% ./speedtest.py --gpu 2
==> Benchmark: gpu=2, times=1000, dynamic_input=False
==> Testing FCN32s with Chainer
Elapsed time: 45.95 [s / 1000 evals]
Hz: 21.76 [hz]
==> Testing FCN32s with PyTorch
Elapsed time: 42.63 [s / 1000 evals]
Hz: 23.46 [hz]
% ./speedtest.py --gpu 3 --dynamic-input
==> Benchmark: gpu=3, times=1000, dynamic_input=True
==> Testing FCN32s with Chainer
Elapsed time: 47.68 [s / 1000 evals]
Hz: 20.97 [hz]
==> Testing FCN32s with PyTorch
Elapsed time: 54.49 [s / 1000 evals]
Hz: 18.35 [hz]
git clone https://github.com/BVLC/caffe.git
cd caffe
cp Makefile.config.example Makefile.config
vim Makefile.config # edit as you like
make -j
make pycaffe
export PYTHONPATH=$(pwd)/python:$PYTHONPATH
cd ..
cd pytorch-fcn
cd examples/voc
./model_caffe_to_pytorch.py