Skip to content

Latest commit

 

History

History
40 lines (29 loc) · 1.63 KB

README.md

File metadata and controls

40 lines (29 loc) · 1.63 KB
Raw
3rd party libraries for evaluations

To evaluate our method, we use DSO for camera pose estimation. DSO is under the GPLv3.0 license. We did slight modifications to the code an a patch is provided. We also use the ScanNet dataset and its data loaders code for both evaluation and training.

This folder includes:

  • A patch code for DSO, that improves robustness on our test data and dumps invalid poses, too.

  • ScanNet decoder, used for loading the rgbd images and camera poses during training and evaluation.

Setup DSO

Assuming all the dependencies of DSO have been installed. To clone, apply the customized changes and make DSO:

sh ./setup_dso.sh

This will build the dso_dataset executable file in dso/build/bin folder.

Setup SensReader

In the SensReader folder:

make

to get the executable sens.

Use SensReader

In the SensReader folder: Suppose the scanNet dataset is saved in SCANNET_PATH and the decoded output folder is OUTPUT_PATH

# decode samples in the training set
python decode.py --dataset_path SCANNET_PATH --output_path OUTPUT_PATH --split_file scannet_train.txt

# decode samples in the validation set
python decode.py --dataset_path SCANNET_PATH --output_path  OUTPUT_PATH --split_file scannet_val.txt

This will decode the .sens files into images with 5 frame intervals.