Link: https://github.com/tanjimin/grad-cam-pytorch-light
A customizable lightweight implementation of Grad-CAM (Gradient-weighted Class Activation Mapping) arXiv. This implementation works for custom models.
from grad_cam import grad_cam
grad_cam(<Model>, <Image>, <Layer>, <Label>)
<Model>: A pytorch model.
<Image>: Transformed image for caculating Grad-CAM, a three dimensional tensor.
<Layer>: The layer to back-prop to for calculating gradients.
<Label>: The label to start back-prop.
See example.py for examples.
'Boxer' label and 'Tiger Cat' label for the same image:
| Boxer | Tiger Cat |
|---|---|
 |
 |
'African elephant, Loxodonta africana' label for an image containing an elephant:
| African elephant, Loxodonta africana |
|---|
 |