simple

Single forward-pass inference over the whole scene.

Classes:

SimpleInferer –

Direct call to the predictor on the whole scene.

SimpleInferer [source] ¶

SimpleInferer(softmax: bool = False)

Bases: Inferer

Direct call to the predictor on the whole scene.

The lightest possible Inferer. Use when the model can consume the entire point cloud in one forward pass (object classification, small scenes).

Parameters:

softmax (bool, default: False ) –

If True, softmax the predictor output over the last dim before returning it. The default returns the predictor output unchanged.

Example

import torch

import torch_pointcloud as tp
from torch_pointcloud.inferers import SimpleInferer

model = tp.create_model("pointnet.s3dis-area5").eval()
data = {
    "pos": torch.rand(4096, 3) * 10,
    "x": torch.rand(4096, 9),
    "batch": torch.zeros(4096, dtype=torch.long),
}

inferer = SimpleInferer()
logits = inferer(data, predictor=lambda d: model(d["x"], d["pos"], d["batch"]))