https://github.com/twuilliam/shrec-sketches-helpers
Helpers for sketch-based 3D shape retrieval experiments (SHREC13, SHREC14, PART-SHREC14)
https://github.com/twuilliam/shrec-sketches-helpers
3d-shapes image-retrieval shrec sketch
Last synced: 21 days ago
JSON representation
Helpers for sketch-based 3D shape retrieval experiments (SHREC13, SHREC14, PART-SHREC14)
- Host: GitHub
- URL: https://github.com/twuilliam/shrec-sketches-helpers
- Owner: twuilliam
- License: mit
- Created: 2020-07-23T14:08:32.000Z (almost 5 years ago)
- Default Branch: master
- Last Pushed: 2020-07-24T17:02:48.000Z (almost 5 years ago)
- Last Synced: 2025-04-11T22:07:21.827Z (21 days ago)
- Topics: 3d-shapes, image-retrieval, shrec, sketch
- Language: Python
- Homepage:
- Size: 81.1 KB
- Stars: 17
- Watchers: 2
- Forks: 2
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# shrec-sketches-helpers
Helper scripts for sketch-based 3D shape experiments.
Useful for rendering and pre-processing SHREC13, SHREC14 and PART-SHREC14 datasets.
## Datasets
SHREC 2013 [[website](http://orca.st.usm.edu/~bli/sharp/sharp/contest/2013/SBR/)]
SHREC 2014 [[website](http://orca.st.usm.edu/~bli/sharp/sharp/contest/2014/SBR/)]
## Image annotations
Run [meta.py](meta.py) to create pandas dataframes for both datasets.
There will be 2 dataframes per dataset: one for the sketches, one for the CAD models.
Each entry in the dataframe consist of the filename path, the label and the split.
## Word vectors
Run [w2v.py](w2v.py) to get the word vector for all class names.
Word vectors are stored in a dictionary in a `.npz` file.
It requires the gensim library:
`conda install -c anaconda gensim`
## Blender 2D rendering
- Download Blender 2.79 [[link](https://download.blender.org/release/)]
- Use the Blender script from the [MVCNN](https://github.com/jongchyisu/mvcnn_pytorch) project
[[link](http://people.cs.umass.edu/~jcsu/papers/shape_recog/render_shaded_black_bg.blend)]
- Render 2D projections with [BlenderPhong](https://github.com/WeiTang114/BlenderPhong)
Following the [MVCNN](http://vis-www.cs.umass.edu/mvcnn/) paper,
render 12 views by placing the camera every 30 degrees, with an elevation of 30 degrees from the ground floor.
This requires to modify the original [`phong.py`](https://github.com/WeiTang114/BlenderPhong/blob/master/phong.py) script.
Check this [fork](https://github.com/twuilliam/BlenderPhong) for more info.
It also requires the python blender package:
`conda install -c kitsune.one python-blender=2.79`
Once everything is installed, run [render.sh](render.sh) to render the 2D views for all CAD models.
## Image pre-processing
### Sketch resizing
Run [resize_sk.py](resize_sk.py). Sketches will be resized to 256x256.
### 2D projection cropping and resizing
Run [resize_cad.py](resize_cad.py).
Images of 1024x1024 will be cropped with a margin of 100px, and resized to 256x256.
## Citation
If you find these scripts useful, please consider citing our paper:
```
@article{
Thong2020OpenSearch,
title={Open Cross-Domain Visual Search},
author={Thong, William and Mettes, Pascal and Snoek, Cees G.M.},
journal={CVIU},
year={2020},
url={https://arxiv.org/abs/1911.08621}
}
```