Skip to content

Latest commit

 

History

History
54 lines (29 loc) · 1.35 KB

VISUALIZE.md

File metadata and controls

54 lines (29 loc) · 1.35 KB
Raw

Visualization Tools

We provide a visualization tool with GradCAM in PyDeepFakeDet, which may facilitate you to better understand the behavior of the pretrained model.

Gradcam

GradCam[1] is a visual explanations for deep networks via gradient-based localization. It uses the gradients of any target category that flow into the final convolutional layer to produce a coarse localization map highlighting important regions in the image.

Requirements

  • PIL

  • matplotlib

Prepare

Before running the code, you need prepare

  • the model you want to visualize, placing in the PyDeepFakeDet/models folder.

  • the checkpoint of the pretrained model

Run

Examples: python visualization/gradcam.py --model Xception --pth Xception.pth --layer res4 --img demo.jpg --save_path save.jpg

args:

optional arguments:
  -h, --help            show this help message and exit
  --model,  model name, should be exactly same with the file name in PyDeepFakeDet/models
  --pth,     checkpoint file of the model
  --img,     facial image path
  --save_path,      save activation map path

Result Demo

Origin Image:

Activation Map:

References

[1] Selvaraju R R, Cogswell M, Das A, et al. Grad-cam: Visual explanations from deep networks via gradient-based localization. in ICCV 2017.