Skip to content

Latest commit

 

History

History
57 lines (26 loc) · 1.76 KB

demo.md

File metadata and controls

57 lines (26 loc) · 1.76 KB
Raw

Demo

Simply run train.py –demo to download the pre-trained models and run on an example airplane image. The output will be saved in PLY format as ./doc/pictures/plane_input_demoAtlasnetReconstruction.ply. You can view the output by downloading [MeshLab] (http://www.meshlab.net/).

Trained models

Trained are automatically downloaded by train.py --demo ....

You can also explicitly download them with:

chmod +x training/download_trained_models.sh
./training/download_trained_models.sh

All training options can be recovered in {dir_name}/options.txt.

  • ./training/trained_models/atlasnet_autoencoder_25_squares/network.pth [Default]

  • ./training/trained_models/atlasnet_autoencoder_1_sphere/network.pth

  • ./training/trained_models/atlasnet_singleview_25_squares/network.pth [Default]

  • ./training/trained_models/atlasnet_singleview_1_sphere/network.pth

Usage

python train.py --demo --demo_input_path YOUR_IMAGE_or_OBJ_PATH --reload_model_path YOUR_MODEL_PTH_PATH 

This function takes an image or pointcloud path as input and save the mesh infered by Atlasnet
Extension supported are `ply` `npy` `obg` and `png`
--demo_input_path input file e.g. image.png or object.ply 
--reload_model_path trained model path (see below for pretrained models) 
:return: path to the generated mesh

To generate the example below, simple run python train.py --demo. It will (1) default to the 2D plane image as input, (2) download a trained single-view Altasnet with 25 square primitives, (3) run the image through the network and (4) save the generated 3D plane in doc/pictures/.

input

You can use our Meshlab Visualization Trick to have nicer visualization of the generated mesh in Meshlab.