saliency
saliency copied to clipboard
PAIR-code
Framework-agnostic implementation for state-of-the-art saliency methods (XRAI, BlurIG, SmoothGrad, and more).
To speed up the computation for obtaining smoothed masks, the noisy samples are created first and gradients are obtained as a single batch for all the noisy samples. Afterwards the...
Thank you for the great repository! From the very cool examples there, I saw that the package supports 2D images. I was wondering if it also offers methods for 3D...
Recent versions of scikit-image have completely deprecated the `selem` argument in favor of `dilation`: [skimage.morphology.dilation](https://scikit-image.org/docs/stable/api/skimage.morphology.html#skimage.morphology.dilation) [Complete deprecation of selem parameter](https://github.com/scikit-image/scikit-image/pull/6583/commits/7430c415ccdf716c1aca7226f03ede85e12e1ba9) This minor patch to [xrai.py](https://github.com/PAIR-code/saliency/blob/master/saliency/core/xrai.py#L126) resolves the issue on line...
In validate_x_y_tensor the expected shape of y is [None] . Does it mean it can only be used for classification task?
Hi PAIR-code/saliency team, Thanks for your great work and contribution to the community! Your package helped a lot for my research. I am just wondering if you would be interested...
Dear PAIR code group, Thank you for sharing this great work. I really like it. I am using the Examples_core.ipynb now. Is there a way to use GPU in this...
Hi, Firstly thanks for this great repo. In the Examples_pytorch.ipynb file, If i change ``call_model_args = {class_idx_str: prediction_class}`` to ``call_model_args = {class_idx_str: }`` XRAI mask still remains that of Doberman...
In your paper, you mention it is possible to implement it on audio. I am confused about how to transfer the audio to the image. Do Ithe plot spectrogram first...
One-line change: adds output_layer = output_layer[:,target_class_idx] to the else statement in def call_model_function. This means that previously Grad-CAM was calculating wrt the entire output layer, this should fix it to...
