# Looking 3D: Anomaly Detection with 2D-3D Alignment

<table>
  <tr>
      <strong><a href="https://openaccess.thecvf.com/content/CVPR2024/papers/Bhunia_Looking_3D_Anomaly_Detection_with_2D-3D_Alignment_CVPR_2024_paper.pdf">Looking 3D: Anomaly Detection with 2D-3D Alignment</a></strong><br>
      Ankan Bhunia, Changjian Li, Hakan Bilen<br>
      CVPR 2024
  </tr>
</table>

[![Website](https://img.shields.io/badge/Project-Website-87CEEB)](https://groups.inf.ed.ac.uk/vico/research/Looking3D)
[![paper](https://img.shields.io/badge/arXiv-Paper-<COLOR>.svg)](https://openaccess.thecvf.com/content/CVPR2024/papers/Bhunia_Looking_3D_Anomaly_Detection_with_2D-3D_Alignment_CVPR_2024_paper.pdf)
[![dataset](https://img.shields.io/badge/Dataset-link-blue)](https://uoe-my.sharepoint.com/:f:/g/personal/s2514643_ed_ac_uk/EjURAFBBbmxHvlMvDGrKvzEBOB29U3QShVRsqekp0rha_g?e=jenLk6)

<img src=figures/title.jpg>

<hr />

## Dataset

- The `BrokenChairs180K` dataset is available for download from [here](https://uoe-my.sharepoint.com/:f:/g/personal/s2514643_ed_ac_uk/EjURAFBBbmxHvlMvDGrKvzEBOB29U3QShVRsqekp0rha_g?e=jenLk6).
- The dataset contains around 180K rendered images with 100K classified as anomaly and 80K normal.
- Each rendered query image is associated with a normal shape reference.
- Different types of abnormalities include: missing parts, broken parts, swapped components, mis-alignments.
- The query pose is unknown.
- Testing is performed on previously unseen instances.

<img src=figures/data_preview.gif>

<table>
  <tr>
    <td><b>filename and download link</b></td>
    <td><b>folder structure</b></td>
    <td><b>size (after extracting)</b></td>
    <td><b>comments</b></td>
  </tr>
      <tr>
    <td><a href="https://uoe-my.sharepoint.com/:u:/g/personal/s2514643_ed_ac_uk/EYoXrt3Ot7ZDmfUNIB9Xq3UBmX5jloND4kW35OaxxLBbTw?e=yieACI">images.zip</a></td>
    <td>BrokenChairs/images/</td>
    <td>21 GB</td>
    <td>[1] (see below)</td>
  </tr>
    <tr>
    <td><a href="https://uoe-my.sharepoint.com/:u:/g/personal/s2514643_ed_ac_uk/Efu7n0UgwZFKoZZzAdl_ccwByIS7af0Ds5D9wQg5SaPAyw?e=U9HtP1">annotations.zip</a></td>
    <td>BrokenChairs/annotations/</td>
    <td>2 GB</td>
      <td>[2] (see below)</td>
  </tr>
      <tr>
    <td><a href="https://uoe-my.sharepoint.com/:u:/g/personal/s2514643_ed_ac_uk/Eda1I8N4eTtIjsomzw2wLsMBiAUSRLmbaKt8QWYD3bGw_Q?e=WrVmeU">shapes.zip</a></td>
    <td>BrokenChairs/shapes/</td>
    <td>14 GB</td>
        <td>[3] (see below)</td>
  </tr>
    <tr>
    <td><a href="https://uoe-my.sharepoint.com/:u:/g/personal/s2514643_ed_ac_uk/Ee6ho7lde_BLqL6fY7V974IBPylEFznzJoNwDLI46qcD-Q?e=fmThG8">split.json</a></td>
    <td>BrokenChairs/split.json</td>
      <td>134 KB</td>
      <td>[4] (see below)</td>
  </tr>
</table>

Note:

[1]`BrokenChairs/images/`: The filenames for the images have a specifc structure. For example in the file with name `render_183_1944_2.5_300_30_3_normal.png`, `183` is the `shape_id`, `1944` is the `texture_id`, `2.5_300_30_3` contains info on camera paramters (in the format of `<radius>_<azim>_<elev>_<light-index>`).

[2]`BrokenChairs/annotations/`:`<info_*>`: It contains 2d_bbox, IoU, camera_parameters and texture_id.
`<mask_new_*>`: binary mask of the object part with the anomaly.
`<mask_old_*>`: binary mask of the object part without the anomaly (normal).
`<mask_new_*>`: segmentation mask of the chair with the anomaly.
`<mask_old_*>`: segmentation mask of the chair without the anomaly (normal).

Annotations are available for anomaly images only. For some anomaly types like missing component, ```<mask_old_*>``` is not available. 

[3]`BrokenChairs/shapes/`: `<mv_images/*.png>`: grayscale multi-view image,
`<mv_images/*.json>`: json file containing intristic and extrinsic parameters of the rendered image,
`<mv_images/*.npy>`: npy file containing 2D-3D correspondence points.
`<model_id.txt>`: corresponding ShapeNet id.

Please refer to `utils/render_multiview.py` which can be used to obtain the above `<png/json/npy>` files from any given `obj/stl/glb` mesh shape.

[4]`BrokenChairs/split.json`: train/test/val split. Each set has mutually exclusive shape instances.

- Distribution of anomaly types within our dataset, categorized by salient chair parts, is shown below.

<img src=figures/part_stats.png width=600px>



## Citation

If you use the results and code for your research, please cite our paper:

```
@article{bhunia2024look3d,
  title={Looking 3D: Anomaly Detection with 2D-3D Alignment},
  author={Bhunia, Ankan Kumar and Li, Changjian and Bilen, Hakan},
  journal={CVPR},
  year={2024}
}
```