Grad-CAM from Scratch with PyTorch Hooks

automobile stops all of the sudden. Worryingly, there is no such thing as a cease sign up sight. The engineers can solely make guesses as to why the automobile’s neural community grew to become confused. It may very well be a tumbleweed rolling throughout the road, a automobile coming down the opposite lane or the pink billboard within the background. To search out the true motive, they flip to Grad-CAM [1].

Grad-CAM is an explainable AI (XAI) method that helps reveal why a convolutional neural community (CNN) made a specific determination. The strategy produces a heatmap that highlights the areas in a picture which can be an important for a prediction. For our self-driving automobile instance, this might present if the pixels from the weed, automobile or billboard induced the automobile to cease.

Now, Grad-CAM is one in all many XAI methods for Computer Vision. Because of its pace, flexibility and reliability, it has rapidly turn out to be probably the most common. It has additionally impressed many associated strategies. So, if you’re involved in XAI, it’s value understanding precisely how this technique works. To try this, we might be implementing Grad-CAM from scratch utilizing Python.

Particularly, we might be counting on PyTorch Hooks. As you will notice, these permit us to dynamically extract gradients and activations from a community throughout ahead and backwards passes. These are sensible abilities that won’t solely permit you to implement Grad-CAM but additionally any gradient-based XAI technique. See the complete challenge on GitHub.

The idea behind Grad-CAM

Earlier than we get to the code, it’s value concerning the speculation behind Grad-CAM. If you need a deep dive, then take a look at the video under. If you wish to study different strategies, then see this free XAI for Computer Vision course.

To summarise, when creating Grad-CAM heatmaps, we begin with a skilled CNN. We then do a ahead cross via this community with a single pattern picture. It will activate all convolutional layers within the community. We name these function maps ($A^ok$). They are going to be a set of 2D matrices that comprise completely different options detected within the pattern picture.

With Grad-CAM, we’re sometimes within the maps from the final convolutional layer of the community. Once we apply the strategy to VGG16, you will notice that its ultimate layer has 512 function maps. We use these as they comprise options with essentially the most detailed semantic data whereas nonetheless retaining spatial data. In different phrases, they inform us what was used for a prediction and the place within the picture it was taken from.

The issue is that these maps additionally comprise options which can be vital for different lessons. To mitigate this, we observe the method proven in Determine 1. As soon as we’ve got the function maps ($A^ok$), we weight them by how vital they’re to the category of curiosity ($y_c$). We do that utilizing $a_k^c$ — the typical gradient of the rating for $y_c$ w.r.t. to the weather within the function map. We then do element-wise summation. For VGG16, you will notice we go from 512 maps of 14×14 pixels to a single 14×14 map.

The gradients for a person ingredient ($frac{partial y^c}{partial A_{ij}^ok}$) inform us how a lot the rating will change with a small change within the ingredient. Which means massive common gradients point out that all the function map was vital and may contribute extra to the ultimate heatmap. So, after we weight and sum the maps, those that comprise options for different lessons will seemingly contribute much less.

The ultimate steps are to use the ReLU activation operate to make sure all destructive parts may have a price of zero. Then we upsample with interpolation so the heatmap has the identical dimensions because the pattern picture. The ultimate map is summarised by the system under. You would possibly recognise it from the Grad-CAM paper [1].

$$ L_{Grad-CAM}^c = ReLUleft( sum_{ok} a_k^c A^ok proper) $$

Grad-CAM from Scratch

Don’t fear if the speculation will not be utterly clear. We are going to stroll via it step-by-step as we apply the strategy from scratch. You could find the complete challenge on GitHub. To start out, we’ve got our imports under. These are all widespread imports for pc imaginative and prescient issues.

import matplotlib.pyplot as plt
import numpy as np

import cv2
from PIL import Picture

import torch
import torch.nn.practical as F
from torchvision import fashions, transforms

import urllib.request

Load pretrained mannequin from PyTorch

We’ll be making use of Grad-CAM to VGG16 pretrained on ImageNet. To assist, we’ve got the 2 features under. The primary will format a picture within the right means for enter into the mannequin. The normalisation values used are the imply and customary deviation of the photographs in ImageNet. The 224×224 measurement can be customary for ImageNet fashions.

def preprocess_image(img_path):

    """Load and preprocess photos for PyTorch fashions."""

    img = Picture.open(img_path).convert("RGB")

    #Transforms utilized by imagenet fashions
    rework = transforms.Compose([
        transforms.Resize((224, 224)),
        transforms.ToTensor(),
        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
    ])

    return rework(img).unsqueeze(0)

ImageNet has many lessons. The second operate will format the output of the mannequin so we show the lessons with the very best predicted chances.

def display_output(output,n=5):

    """Show the highest n classes predicted by the mannequin."""
    
    # Obtain the classes
    url = "https://uncooked.githubusercontent.com/pytorch/hub/grasp/imagenet_classes.txt"
    urllib.request.urlretrieve(url, "imagenet_classes.txt")

    with open("imagenet_classes.txt", "r") as f:
        classes = [s.strip() for s in f.readlines()]

    # Present prime classes per picture
    chances = torch.nn.practical.softmax(output[0], dim=0)
    top_prob, top_catid = torch.topk(chances, n)

    for i in vary(top_prob.measurement(0)):
        print(classes[top_catid[i]], top_prob[i].merchandise())

    return top_catid[0]

We now load the pretrained VGG16 model (line 2), transfer it to a GPU (traces 5-8) and set it to analysis mode (line 11). You may see a snippet of the mannequin output in Determine 2. VGG16 is product of 16 weighted layers. Right here, you possibly can see the final 2 of 13 convolutional layers and the three totally related layers.

# Load the pre-trained mannequin (e.g., VGG16)
mannequin = fashions.vgg16(pretrained=True)

# Set the mannequin to gpu
machine = torch.machine('mps' if torch.backends.mps.is_built() 
                      else 'cuda' if torch.cuda.is_available() 
                      else 'cpu')
mannequin.to(machine)

# Set the mannequin to analysis mode
mannequin.eval()

The names you see in Determine 2 are vital. Later, we are going to use them to reference a selected layer within the community to entry its activations and gradients. Particularly, we are going to use mannequin.options[28]. That is the ultimate convolutional layer within the community. As you possibly can see within the snapshot, this layer comprises 512 function maps.

Ahead cross with pattern picture

We might be explaining a prediction from this mannequin. To do that, we want a pattern picture that might be fed into the mannequin. We downloaded one from Wikipedia Commons (traces 2-3). We then load it (traces 5-6), crop it to have equal top and width (line 7) and show it (traces 9-10). In Determine 3, you possibly can see we’re utilizing a picture of a whale shark in an aquarium.

# Load a pattern picture from the online
img_url = "https://add.wikimedia.org/wikipedia/commons/thumb/a/a1/Male_whale_shark_at_Georgia_Aquarium.jpg/960px-Male_whale_shark_at_Georgia_Aquarium.jpg"
urllib.request.urlretrieve(img_url, "sample_image.jpg")[0]

img_path = "sample_image.jpg"
img = Picture.open(img_path).convert("RGB")
img = img.crop((320, 0, 960, 640))  # Crop to 640x640

plt.imshow(img)
plt.axis("off")

ImageNet has no devoted class for whale sharks, so will probably be fascinating to see what the mannequin predicts. To do that, we begin by processing our picture (line 2) and shifting it to the GPU (line 3). We then do a ahead cross to get a prediction (line 6) and show the highest 5 chances (line 7). You may see these in Determine 4.

# Preprocess the picture
img_tensor = preprocess_image(img_path)
img_tensor = img_tensor.to(machine)

# Ahead cross
predictions = mannequin(img_tensor)
display_output(predictions,n=5)

Given the accessible lessons, these appear affordable. They’re all marine life and the highest two are sharks. Now, let’s see how we will clarify this prediction. We need to perceive what areas of the picture contribute essentially the most to the very best predicted class — hammerhead.

PyTorch hooks naming conventions

Grad-CAM heatmaps are created utilizing each activations from a ahead cross and gradients from a backwards cross. To entry these, we are going to use PyTorch hooks. These are features that permit you to save the inputs and outputs of a layer. We received’t do it right here, however they even permit you to alter these facets. For instance, Guided Backpropagation could be utilized by making certain solely optimistic gradients are propagated utilizing a backwards hook.

You may see some examples of those features under. A forwards_hook might be known as throughout a ahead cross. It will likely be registered on a given module (i.e. layer). By default, the operate receives three arguments — the module, its enter and its output. Equally, a backwards_hook is triggered throughout a backwards cross with the module and gradients of the enter and output.

# Instance of a forwards hook operate
def fowards_hook(module, enter, output):
    """Parameters:
            module (nn.Module): The module the place the hook is utilized.
            enter (tuple of Tensors): Enter to the module.
            output (Tensor): Output of the module."""
    ...

# Instance of a backwards hook operate 
def backwards_hook(module, grad_in, grad_out):
    """Parameters:
            module (nn.Module): The module the place the hook is utilized.
            grad_in (tuple of Tensors): Gradients w.r.t. the enter of the module.
            grad_out (tuple of Tensors): Gradients w.r.t. the output of the module."""
    ...

To keep away from confusion, let’s make clear the parameter names utilized by these features. Check out the overview of the usual backpropagation process for a convolutional layer in Determine 5. This layer consists of a set of kernels, $Okay$, and biases, $b$. The opposite elements are the:

• enter – a set of function maps or a picture
• output – set of function maps
• grad_in is the gradient of the loss w.r.t. the layer’s enter.
• grad_out is the gradient of the loss w.r.t. the layer’s output.

We have now labelled these utilizing the identical names of the arguments used to name the hook features that we apply later.

Be mindful, we received’t use the gradients in the identical means as backpropagation. Often, we use the gradients of a batch of photos to replace $Okay$ and $b$. Now, we’re solely involved in grad_out of a single pattern picture. It will give us the gradients of the weather within the layer’s function maps. In different phrases, the gradients we use to weight the function maps.

Activations with PyTorch ahead hook

Our VGG16 community has been created utilizing ReLU with inplace=True. These modify tensors in reminiscence, so the unique values are misplaced. That’s, tensors used as enter are overwritten by the ReLU operate. This will result in issues when making use of hooks, as we may have the unique enter. So we use the code under to interchange all ReLU features with inplace=False ones. This is not going to influence the output of the mannequin, however it should enhance its reminiscence utilization.

# Exchange all in-place ReLU activations with out-of-place ones
def replace_relu(mannequin):

    for identify, baby in mannequin.named_children():
        if isinstance(baby, torch.nn.ReLU):
            setattr(mannequin, identify, torch.nn.ReLU(inplace=False))
            print(f"Changing ReLU activation in layer: {identify}")
        else:
            replace_relu(baby)  # Recursively apply to submodules

# Apply the modification to the VGG16 mannequin
replace_relu(mannequin)

Beneath we’ve got our first hook operate — save_activations. It will append the output from a module (line 6) to a listing of activations (line 2). In our case, we are going to solely register the hook onto one module (i.e. the final convolutional layer), so this checklist will solely comprise one ingredient. Discover how we format the output (line 6). We detach it from the computational graph so the community will not be affected. We additionally format them as a numpy array and squeeze the batch dimension.

# Record to retailer activations
activations = []

# Operate to avoid wasting activations
def save_activations(module, enter, output):
    activations.append(output.detach().cpu().numpy().squeeze())

To make use of the hook operate, we register it on the final convolutional layer — mannequin.options[28]. That is accomplished utilizing the register_forward_hook operate.

# Register the hook to the final convolutional layer
hook = mannequin.options[28].register_forward_hook(save_activations)

Now, after we do a ahead cross (line 2), the save_activations hook operate might be known as for this layer. In different phrases, its output might be saved to the activations checklist.

# Ahead cross via the mannequin to get activations
prediction = mannequin(img_tensor)

Lastly, it’s good follow to take away the hook operate when it’s not wanted (line 2). This implies the ahead hook operate is not going to be triggered if we do one other ahead cross.

# Take away the hook after use
hook.take away()  

The form of those activations is (512, 14, 14). In different phrases, we’ve got 512 function maps and every map is 14×14 pixels. You may see some examples of those in Determine 6. A few of these maps could comprise options vital for different lessons or those who lower the chance of the anticipated class. So let’s see how we will discover gradients to assist establish an important maps.

act_shape = np.form(activations[0])
print(f"Form of activations: {act_shape}") # (512, 14, 14)

Gradients with PyTorch backwards hooks

To get gradients, we observe an analogous course of to earlier than. The important thing distinction is that we now use the register_full_backward_hook to register the save_gradients operate (line 7). It will be sure that it’s known as throughout a backwards cross. Importantly, we do the backwards cross (line 16) from the output for the category with the very best rating (line 13). This successfully units the rating for this class to 1 and all different scores to 0. In different phrases, we get the gradients of the hammerhead class w.r.t. to the weather of the function maps.

gradients = []

def save_gradient(module, grad_in, grad_out):
    gradients.append(grad_out[0].cpu().numpy().squeeze())

# Register the backward hook on a convolutional layer
hook = mannequin.options[28].register_full_backward_hook(save_gradient)

# Ahead cross
output = mannequin(img_tensor)

# Choose the category with highest rating
rating = output[0].max()

# Backward cross from the rating
rating.backward()

# Take away the hook after use
hook.take away()

We may have a gradient for each ingredient of the function maps. So, once more, the form is (512, 14, 14). Determine 7 visualises a few of these. You may see some are likely to have larger values. Nonetheless, we’re not so involved with the person gradients. Once we create a Grad-CAM heatmap, we are going to use the typical gradient of every function map.

grad_shape = np.form(gradients[0])
print(f"Form of gradients: {grad_shape}") # (512, 14, 14)

Lastly, earlier than we transfer on, it’s good follow to reset the mannequin’s gradients (line 2). That is significantly vital if you happen to plan to run the code for a number of photos, as gradients could be collected with every backwards cross.

# Reset gradients
mannequin.zero_grad() 

Creating Grad-CAM heatmaps

First, we discover the imply gradients for every function map. There might be 512 of those common gradients. Plotting a histogram of them, you possibly can see most are typically round 0. In different phrases, these don’t have a lot influence on the anticipated rating. There are a number of that are likely to have a destructive influence and a optimistic influence. It’s these function maps we need to give extra weight to.

# Step 1: combination the gradients
gradients_aggregated = np.imply(gradients[0], axis=(1, 2))

We mix all of the activations by doing element-wise summation (traces 2-4). Once we do that, we weight every function map by its common gradient (line 3). In the long run, we may have one 14×14 array.

# Step 2: weight the activations by the aggregated gradients and sum them up
weighted_activations = np.sum(activations[0] * 
                              gradients_aggregated[:, np.newaxis, np.newaxis], 
                              axis=0)

These weighted activations will comprise each optimistic and destructive pixels. We will contemplate the destructive pixels to be suppressing the anticipated rating. In different phrases, a rise within the worth of those areas tends to lower the rating. Since we’re solely within the optimistic contributions—areas that assist the category prediction—we apply a ReLU activation to the ultimate heatmap (line 2). You may see the distinction within the heatmaps in Determine 9.

# Step 3: ReLU summed activations
relu_weighted_activations = np.most(weighted_activations, 0)

You may see the heatmap in Determine 9 is kind of coarse. It might be extra helpful if it had the size of the unique picture. For this reason the final step for creating Grad-CAM heatmaps is to upsample to the dimension of the enter picture (traces 2-4). On this case, we’ve got a 224×224 picture.

#Step 4: Upsample the heatmap to the unique picture measurement
upsampled_heatmap = cv2.resize(relu_weighted_activations, 
                               (img_tensor.measurement(3), img_tensor.measurement(2)), 
                               interpolation=cv2.INTER_LINEAR)

print(np.form(upsampled_heatmap))  # Needs to be (224, 224)

Determine 10 offers us our ultimate visualisation. We show the pattern picture (traces 5-7) subsequent to the heatmap (traces 10-15). For the latter, we create a transparent visualisation with the assistance of Canny Edge detection (line 10). This provides us an edge map (i.e. define) of the pattern picture. We will then overlay the heatmap on prime of this (line 14).

# Step 5: visualise the heatmap
fig, ax = plt.subplots(1, 2, figsize=(8, 8))

# Enter picture
resized_img = img.resize((224, 224))
ax[0].imshow(resized_img)
ax[0].axis("off")

# Edge map for the enter picture
edge_img = cv2.Canny(np.array(resized_img), 100, 200)
ax[1].imshow(255-edge_img, alpha=0.5, cmap='grey')

# Overlay the heatmap 
ax[1].imshow(upsampled_heatmap, alpha=0.5, cmap='coolwarm')
ax[1].axis("off")

our Grad-CAM heatmap, there may be some noise. Nonetheless, it seems the mannequin is counting on the tail fin and, to a lesser extent, the pectoral fin to make its predictions. It’s beginning to make sense why the mannequin categorised this shark as a hammerhead. Maybe each animals share these traits.

For some additional investigation, we apply the identical course of however now utilizing an precise picture of a hammerhead. On this case, the mannequin seems to be counting on the identical options. It is a bit regarding. Would we not anticipate the mannequin to make use of one of many shark’s defining options— the hammerhead? In the end, this may increasingly lead VGG16 to confuse various kinds of sharks.

With this instance, we see how Grad-CAM can spotlight potential flaws in our mannequin. We cannot solely get their predictions but additionally perceive how they made them. We will perceive if the options used will result in unexpected predictions down the road. This will doubtlessly save us loads of time, cash and within the case of extra consequential purposes, lives!

If you wish to study extra about XAI for CV take a look at one in all these articles. Or see this Free XAI for CV course.