Dataset Preparation

Step 1. Rename the Virtual-Home output directory to ‘vh.[name]’, for example: ‘vh.door’ Make sure the directory contains 100+ frames of frame_id-camera_id-point_cloud.exr, frame_id.json, and frame_id-camera_id-rgb.png.

Step 2. Convert to dataset. The first 100 frames are used for training and the rest for evaluation.

python [name]

This generates vh.[name].train.pth and vh.[name].eval.pth.


python [name]


Specify the frame id [eval_id] for evaluation.

python [name] [eval_id]


View Github