Notice

  1. Use CUE as baseline.

Ideas

  1. Use information theory to explain the process.

Plan

  1. Use uncertainty estimation method in CUE.
    1. Evaluate CUE on semantic kitti (outdoor), or scannet v2 (indoor).
    2. Extract uncertainty estimation method of CUE.
  2. Achieve socket communication for model inference for CUE model.
    1. Socket communication.
    2. Local small model training, calculate uncertainty information.
    3. Cloud large model training, with global feature map and compressed point cloud input.
  3. Deploy to edge device.
    1. Select proper device.
  4. Evaluate performance in simulated setting.
    1. Latency.
    2. Accuracy.

Question

Why use outdoor data, instead of indoor data?
In door first, everything. Cloud dummy model.
The server disk is full, no space left.
Waiting for free space. Today.
What suppose to be the edge device? Is raspberry pi sounds good?
I have a respberry pi, and can use it to do the task. xavier, arin, raspberry pi, desktop as edge device.