TestStep.md

We recommend you to directly compare with the released results.

Step 1 Generate mask

cd ./dynamic
run ./scripts/test_cityscapes.sh to generate dynamic masks, which indicates the moving cars

use ./process_scripts/gen_nonrigid_small/non_rigid.py to generate non rigid masks(including person and biker)

use ./process_scripts/gen_nonrigid_small/small.py to generate small object masks.

Step 2 Background prediction

./back/scripts/street/test.sh

Step 3 Background Inpainting

We use the Generative Inpainting to inpaint missing region

Step 4 Dynamic Motion Prediction

use the python files
./process_scripts/test_cityscapes.py ./process_scripts/test_kitti.py
to generate list storing the paths for testing

Then run
./fore/script/test_city.sh

Step 5 Compute Occlusion Map

use the script ./process_scripts/occ.py

Step 6 Video Inpainting

We use the Deep-Flow-Guided-Video-Inpainting to inpaint occlusion area.

Please note that, for Cityscapes dataset, we run the test procedure twice. After finishing the prediction of next 5 frames, we run the semantic segmenation method on predicted frames to obtain their semantic maps. It is used for next 5 to 10 frames prediction