Simple way to leverage the class-specific activation of convolutional layers in PyTorch.
Source: image from woopets (activation maps created with a pretrained Resnet-18)
TorchCAM leverages PyTorch hooking mechanisms to seamlessly retrieve all required information to produce the class activation without additional efforts from the user. Each CAM object acts as a wrapper around your model.
You can find the exhaustive list of supported CAM methods in the documentation, then use it as follows:
# Define your model
from torchvision.models import get_model, get_model_weights
model = get_model("resnet18", weights=get_model_weights("resnet18").DEFAULT).eval()
# Set your CAM extractor
from torchcam.methods import LayerCAM
cam_extractor = LayerCAM(model)Please note that by default, the layer at which the CAM is retrieved is set to the last non-reduced convolutional layer. If you wish to investigate a specific layer, use the target_layer argument in the constructor.
Once your CAM extractor is set, you only need to use your model to infer on your data as usual. If any additional information is required, the extractor will get it for you automatically.
from torchvision.io import decode_image
from torchvision.transforms.functional import normalize, resize, to_pil_image
from torchvision.models import get_model, get_model_weights
from torchcam.methods import LayerCAM
weights = get_model_weights("resnet18").DEFAULT
model = get_model("resnet18", weights=weights).eval()
preprocess = weights.transforms()
# Get your input
img = decode_image("path/to/your/image.png")
# Preprocess it for your chosen model
input_tensor = preprocess(img)
with LayerCAM(model) as cam_extractor:
# Preprocess your data and feed it to the model
out = model(input_tensor.unsqueeze(0))
# Retrieve the CAM by passing the class index and the model output
activation_map = cam_extractor(out.squeeze(0).argmax().item(), out)If you want to visualize your heatmap, you only need to cast the CAM to a numpy ndarray:
import matplotlib.pyplot as plt
# Visualize the raw CAM
plt.imshow(activation_map[0].squeeze(0).numpy()); plt.axis('off'); plt.tight_layout(); plt.show()
Or if you wish to overlay it on your input image:
import matplotlib.pyplot as plt
from torchcam.utils import overlay_mask
# Resize the CAM and overlay it
result = overlay_mask(to_pil_image(img), to_pil_image(activation_map[0].squeeze(0), mode='F'), alpha=0.5)
# Display it
plt.imshow(result); plt.axis('off'); plt.tight_layout(); plt.show()
Python 3.11 (or higher) and uv/pip are required to install TorchCAM.
You can install the last stable release of the package using pypi as follows:
pip install torchcamAlternatively, if you wish to use the latest features of the project that haven't made their way to a release yet, you can install the package from source:
git clone https://github.com/frgfm/torch-cam.git
pip install -e torch-cam/.This project is developed and maintained by the repo owner, but the implementation was based on the following research papers:
- Learning Deep Features for Discriminative Localization: the original CAM paper
- Grad-CAM: GradCAM paper, generalizing CAM to models without global average pooling.
- Grad-CAM++: improvement of GradCAM++ for more accurate pixel-level contribution to the activation.
- Smooth Grad-CAM++: SmoothGrad mechanism coupled with GradCAM.
- Score-CAM: score-weighting of class activation for better interpretability.
- SS-CAM: SmoothGrad mechanism coupled with Score-CAM.
- IS-CAM: integration-based variant of Score-CAM.
- XGrad-CAM: improved version of Grad-CAM in terms of sensitivity and conservation.
- Layer-CAM: Grad-CAM alternative leveraging pixel-wise contribution of the gradient to the activation.
Source: YouTube video (activation maps created by Layer-CAM with a pretrained ResNet-18)
The full package documentation is available here for detailed specifications.
A minimal demo app is provided for you to play with the supported CAM methods! Feel free to check out the live demo on
If you prefer running the demo by yourself, you will need an extra dependency (Streamlit) for the app to run:
pip install -e ".[demo]"
You can then easily run your app in your default browser by running:
streamlit run demo/app.py
An example script is provided for you to benchmark the heatmaps produced by multiple CAM approaches on the same image:
python scripts/cam_example.py --arch resnet18 --class-idx 232 --rows 2
All script arguments can be checked using python scripts/cam_example.py --help
The purpose of CAM methods is to provide interpretability and they do so by pointing the biggest influence factors on the model outputs. Ideally the CAM should pinpoint all the visual cues that have any influence of the output classification score. For this, we use two metrics:
- Increase in Confidence (higher is better): if we forward the input masked with the CAM (keep origin pixel values where CAM is highest, nullify where lowest), how many times in the dataset has the classification probability improve.
- Average Drop (lower is better): if we forward the input masked with the CAM (keep origin pixel values where CAM is highest, nullify where lowest), by how much does the classification probability drop.
| CAM method | Arch | Average drop (↓) | Increase in confidence (↑) |
|---|---|---|---|
| GradCAM | resnet18 | 0.2686 | 0.2250 |
| GradCAMpp | resnet18 | 0.5271 | 0.1962 |
| SmoothGradCAMpp | resnet18 | 0.2088 | 0.2499 |
| LayerCAM | resnet18 | 0.1712 | 0.2819 |
| GradCAM | mobilenet_v3_large | 0.2678 | 0.3483 |
| GradCAMpp | mobilenet_v3_large | 0.3182 | 0.2535 |
| SmoothGradCAMpp | mobilenet_v3_large | 0.2681 | 0.2678 |
| LayerCAM | mobilenet_v3_large | 0.2526 | 0.2882 |
This benchmark was performed over the validation set of imagenette, which is a subset of Imagenet, on (224, 224) inputs.
You can run this performance benchmark for any CAM method on your hardware as follows:
python scripts/eval_perf.py ~/Downloads/imagenette LayerCAM --arch mobilenet_v3_largeAll script arguments can be checked using python scripts/eval_perf.py --help
You crave for beautiful activation maps, but you don't know whether it fits your needs in terms of latency?
In the table below, you will find a latency overhead benchmark (forward pass not included) for all CAM methods:
| CAM method | Arch | GPU mean (std) | CPU mean (std) |
|---|---|---|---|
| CAM | resnet18 | 0.11ms (0.02ms) | 0.14ms (0.03ms) |
| GradCAM | resnet18 | 3.71ms (1.11ms) | 40.66ms (1.82ms) |
| GradCAMpp | resnet18 | 5.21ms (1.22ms) | 41.61ms (3.24ms) |
| SmoothGradCAMpp | resnet18 | 33.67ms (2.51ms) | 239.27ms (7.85ms) |
| ScoreCAM | resnet18 | 304.74ms (11.54ms) | 6796.89ms (415.14ms) |
| XGradCAM | resnet18 | 3.78ms (0.96ms) | 40.63ms (2.03ms) |
| LayerCAM | resnet18 | 3.65ms (1.04ms) | 40.91ms (1.79ms) |
| CAM | mobilenet_v3_large | N/A* | N/A* |
| GradCAM | mobilenet_v3_large | 8.61ms (1.04ms) | 26.64ms (3.46ms) |
| GradCAMpp | mobilenet_v3_large | 8.83ms (1.29ms) | 25.50ms (3.10ms) |
| SmoothGradCAMpp | mobilenet_v3_large | 77.38ms (3.83ms) | 156.25ms (4.89ms) |
| ScoreCAM | mobilenet_v3_large | 35.19ms (2.11ms) | 679.16ms (55.04ms) |
| XGradCAM | mobilenet_v3_large | 8.41ms (0.98ms) | 24.21ms (2.94ms) |
| LayerCAM | mobilenet_v3_large | 8.02ms (0.95ms) | 25.14ms (3.17ms) |
*The base CAM method cannot work with architectures that have multiple fully-connected layers
This benchmark was performed over 100 iterations on (224, 224) inputs, on a laptop to better reflect performances that can be expected by common users. The hardware setup includes an Intel(R) Core(TM) i7-10750H for the CPU, and a NVIDIA GeForce RTX 2070 with Max-Q Design for the GPU.
You can run this latency benchmark for any CAM method on your hardware as follows:
python scripts/eval_latency.py SmoothGradCAMppAll script arguments can be checked using python scripts/eval_latency.py --help
Looking for more illustrations of TorchCAM features? You might want to check the Jupyter notebooks designed to give you a broader overview.
If you wish to cite this project, feel free to use this BibTeX reference:
@misc{torcham2020,
title={TorchCAM: class activation explorer},
author={François-Guillaume Fernandez},
year={2020},
month={March},
publisher = {GitHub},
howpublished = {\url{https://github.com/frgfm/torch-cam}}
}Feeling like extending the range of possibilities of CAM? Or perhaps submitting a paper implementation? Any sort of contribution is greatly appreciated!
You can find a short guide in CONTRIBUTING to help grow this project!
Distributed under the Apache 2.0 License. See LICENSE for more information.
