Inference on a single unlabeled .ply file about superpoint_transformer HOT 1 CLOSED

Hello @faridht,

Normally you should be able to find what you need for running the inference on a cloud in notebooks/demo.ipynb. This notebook shows how to run a forward pass on a NAG and visualize it with our tool.

Yet, since preparing a NAG for inference requires some preprocessing, you indeed should implement dedicated classes inheriting from BaseDataset and BaseDatamodule for your new data. If you have only one cloud tile in your dataset, this may sound overkill, but this is the cleanest, foolproof way of accessing all the features of this codebase.

A workaround would be that you manually instantiate and run the series of Transform for processing your raw data into a proper NAG for inference, and then run a forward pass on the model. These would be the ones you would otherwise find in your dataset's config file:

• pre_transform
• val_transform
• on_device_val_transform

If you are familiar enough with the project and hydra in particular, you definitely can try this strategy ! However, I will not dedicate much time further helping with this approach, since we do not officially support it and I foresee doing so would open the door to a whole world of new issues 😅

from superpoint_transformer.