Python script for performing depth completion from sparse depth and rgb images using the msg_chn_wacv20. model in Tensorflow Lite.

• OpenCV and tensorflow or tflite_runtime. Also, unrealcv is only required if you want to generate new data using unrealcv.

For the tflite runtime, you can either use tensorflow pip install tensorflow or the TensorFlow Runtime binary

The input images and depth are generated using the UnrealCV library (https://unrealcv.org/), you can find more information about how to generate this data in this other repository for Unreal Synthetic depth generation.

pip install -r requirements.txt

The original models were converted to different formats (including .tflite) by PINTO0309, download the models from his repository and save them into the models folder.

The Pytorch pretrained model was taken from the original repository.

For ONNX inference, use this other repository:https://github.com/ibaiGorordo/ONNX-msg_chn_wacv20-depth-completion

• Video inference (UnrealCV synthetic data):

python video_depth_estimation.py

• msg_chn_wacv20 model: https://github.com/anglixjtu/msg_chn_wacv20

• PINTO0309's model zoo: https://github.com/PINTO0309/PINTO_model_zoo

• PINTO0309's model conversion tool: https://github.com/PINTO0309/openvino2tensorflow

• Original paper: https://openaccess.thecvf.com/content_WACV_2020/papers/Li_A_Multi-Scale_Guided_Cascade_Hourglass_Network_for_Depth_Completion_WACV_2020_paper.pdf