p61049258
/
FCA
forked from idrl/FCA
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Merge branch 'gh-pages' of github.com:winterwindwang/Full-coverage-camouflage-adversarial-attack into gh-pages

This commit is contained in:
freeneuro 2021-09-12 17:29:33 +08:00
parent 8b4dc10928 fe964ff0fa
commit 8ab7be504c
1 changed files with 60 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

62
README.md
View File

@ -1,2 +1,60 @@
# Full-coverage-camouflage-adversarial-attack
## code and example are public available!
## FCA: Learning a 3D Full-coverage Vehicle Camouflage for Multi-view Physical Adversarial Attack
Case study of the FCA. The code can be find in [FCA](https://github.com/winterwindwang/Full-coverage-camouflage-adversarial-attack.git).
### Cases of digital attack
#### Carmear distance at 3
<table>
<tr>
<td><center> <img src = 'https://github.com/winterwindwang/Full-coverage-camouflage-adversarial-attack/blob/gh-pages/assets/distance_3_elevation_0_ori_pred.gif?raw=true'/></center></td>
<td><center><img src = 'https://github.com/winterwindwang/Full-coverage-camouflage-adversarial-attack/blob/gh-pages/assets/distance_3_elevation_0_adv_pred.gif?raw=true'/></center></td>
</tr>
<tr>
<td><center> <img src = 'https://github.com/winterwindwang/Full-coverage-camouflage-adversarial-attack/blob/gh-pages/assets/distance_3_elevation_30_ori_pred.gif?raw=true'/></center></td>
<td><center><img src = 'https://github.com/winterwindwang/Full-coverage-camouflage-adversarial-attack/blob/gh-pages/assets/distance_3_elevation_30_adv_pred.gif?raw=true'/></center></td>
</tr>
<tr>
<td><center> <img src = 'https://github.com/winterwindwang/Full-coverage-camouflage-adversarial-attack/blob/gh-pages/assets/distance_3_elevation_50_ori_pred.gif?raw=true'/></center></td>
<td><center><img src = 'https://github.com/winterwindwang/Full-coverage-camouflage-adversarial-attack/blob/gh-pages/assets/distance_3_elevation_50_adv_pred.gif?raw=true'/></center></td>
</tr>
</table>
#### Carmear distance at 5
<table>
<tr>
<td><center> <img src = 'https://github.com/winterwindwang/Full-coverage-camouflage-adversarial-attack/blob/gh-pages/assets/distance_5_elevation_20_ori_pred.gif?raw=true'/></center></td>
<td><center><img src = 'https://github.com/winterwindwang/Full-coverage-camouflage-adversarial-attack/blob/gh-pages/assets/distance_5_elevation_20_adv_pred.gif?raw=true'/></center></td>
</tr>
<tr>
<td><center> <img src = 'https://github.com/winterwindwang/Full-coverage-camouflage-adversarial-attack/blob/gh-pages/assets/distance_5_elevation_50_ori_pred.gif?raw=true'/></center></td>
<td><center><img src = 'https://github.com/winterwindwang/Full-coverage-camouflage-adversarial-attack/blob/gh-pages/assets/distance_5_elevation_50_adv_pred.gif?raw=true'/></center></td>
</tr>
</table>
#### Carmear distance at 10
<table>
<tr>
<td><center> <img src = 'https://github.com/winterwindwang/Full-coverage-camouflage-adversarial-attack/blob/gh-pages/assets/distance_10_elevation_30_ori_pred.gif?raw=true'/></center></td>
<td><center><img src = 'https://github.com/winterwindwang/Full-coverage-camouflage-adversarial-attack/blob/gh-pages/assets/distance_10_elevation_30_adv_pred.gif?raw=true'/></center></td>
</tr>
<tr>
<td><center> <img src = 'https://github.com/winterwindwang/Full-coverage-camouflage-adversarial-attack/blob/gh-pages/assets/distance_10_elevation_50_ori_pred.gif?raw=true'/></center></td>
<td><center><img src = 'https://github.com/winterwindwang/Full-coverage-camouflage-adversarial-attack/blob/gh-pages/assets/distance_10_elevation_50_adv_pred.gif?raw=true'/></center></td>
</tr>
</table>
### Cases of multi-view robust
### Ablation study
#### Different combination of loss terms
<img src = 'https://github.com/winterwindwang/Full-coverage-camouflage-adversarial-attack/blob/gh-pages/assets/abaltion_study_loss.png?raw=true'/>
As we can see from the Figure, different loss term plays different role in attacking. For example, the camouflaged car generated by `obj+smooth (we omit the smooth loss, and denotes as obj)` hardly hidden from the detector, while the camouflaged car generated by `iou` successfully suppress the detecting bounding box of the car region, and finally the camouflaged car generated by `cls` successfully make the detector to misclassify the car to anther category.