Install Tensorflow v2.x and Anaconda libraries. The code was run in python 3.6 and GPU is expected but not required. The codes are borrowed from Here
Download and pre-process the scannet data following instructions in the Scannet directory.
To get axis aligned bounding boxes for Scannet data, refer to this repo. Project the boxes to the rgb images and annotate the heading angles for the boxes in the scenes. Follow guide on box_projection notebook to project and annotate the boxes.
Run python data_pipeline.py to create tensorflow examples for the data. Train the model by running python train.py from the pose_depth_estimation directory. After training, run python eval.py to evalute the model. The codes for training and evaluation codes were adapted from Here