For all of our experiments in the main text with open datasets, we provide source code to reproduce the results. This includes figure. 3 to figure. 5 in the main paper, and various results in the Appendix, with minimal modification to the code.
Platform: Linux
GPU (tested on): NVIDIA GeForce RTX 2080 Ti
CPU (tested on): Our anonymous computing cluster
We recommend using you own python3 virtual environment.
Just create and activate your own virtual environment,
mkdir ~/.venv 2>dev/null
pushd ~/.venv
python3 -m venv --system-site-packages MyEnvName
# alter the VE's name as you like
cd MyEnvName
source bin/activate
popd
python3 -m pip install --upgrade pip setuptools wheel
then install the dependency specified by requirements.txt:
python3 -m pip install -r requirements.txt.
Then you are all set!
-
Reproducing figure. 3 (a), (b), (d), (e) and figure. 4, see FCN/
-
Reproducing figure. 3 (c), (f), see ResNet50/
-
Reproducing figure. 5, see Temporal/
1 and 3 can be run on single CPU in short time, while 2 preferably be run on GPU.