PointCNN_Pytorch

This repo has (**adali PCNN train (2ways) + visualise inference (whole rooms not normed blocks))

Environment:

Run the yml file inside conda to create a separate environment with all dependencies. The command itself can also be found inside the yml.

Run: what is what

run.py is the main file. v1.zip/v2.zip are the latest version of the base repo with same file structure with updated scripts Generic name in this sereis: v(N).zip

v1_txt_ dataloader.zip (input diversity)

load data directly from txt bypassing the h5,npy intermediary. Generic name in this sereis: v(N)txt dataloader.zip

v2_visualize.zip

get whole room predictions and unnormalized visualisatoin not per block normed squashed visuals. run_visualize3D.py inside is the main script to run. However, data_class_unnormalized.py is the script that has all the core functinos to read only the unique blocks inside a room in a non overlapping way. When using txt dataloader directly, the path in the config file should be to the raw Stanford3dDataset_v1.2_Aligned_Version folder with area subfolders, not any txt file with explicit h5 paths as in normal dataloader. Generic name in this sereis: v(N)_visualize.zip

inference.py inside zips [v4 and onwards inference.py correct]

This script is used for loading a pretrained model and run inference to output eval metrics

tahmid0007 / PointCNN_Pytorch