torch-anatomy 0.1.0__tar.gz

torch_anatomy-0.1.0/LICENSE

@@ -0,0 +1,5 @@
+MIT License
+
+Copyright (c) 2024 <Your Name>
+
+Permission is hereby granted, free of charge, to any person obtaining a copy...

torch_anatomy-0.1.0/PKG-INFO

@@ -0,0 +1,18 @@
+Metadata-Version: 2.4
+Name: torch-anatomy
+Version: 0.1.0
+Summary: Layer-by-layer visualizer for PyTorch models
+Author: Harshal Vilas Kale
+License: MIT
+License-File: LICENSE
+Requires-Dist: torch
+Requires-Dist: torchvision
+Requires-Dist: matplotlib
+Requires-Dist: numpy
+Requires-Dist: Pillow
+Requires-Dist: click
+Dynamic: author
+Dynamic: license
+Dynamic: license-file
+Dynamic: requires-dist
+Dynamic: summary

torch_anatomy-0.1.0/README.md

@@ -0,0 +1,43 @@
+# torch-anatomy
+
+**Layer-by-layer visualizer for PyTorch models — Understand what each layer actually does.**
+
+![PyPI](https://img.shields.io/pypi/v/torch-anatomy)
+![License](https://img.shields.io/github/license/yourusername/torch-anatomy)
+
+## Install
+
+```bash
+pip install torch-anatomy
+```
+
+## Usage
+
+```python
+from torch_anatomy import visualize_layers
+from torchvision import models
+
+model = models.resnet18(pretrained=True)
+visualize_layers(
+    model=model,
+    input_image='dog.jpg',
+    layers_to_show=['Conv2d', 'ReLU'],
+    channels_per_layer=6,
+    colormap='inferno',
+    show_colorbar=True
+)
+```
+
+Or from CLI:
+
+```bash
+torch-anatomy --model resnet18 --image dog.jpg
+```
+
+## Features
+- Plug-and-play for any PyTorch CNN
+- Visualizes feature maps for key layers
+- Customizable channels, colormap, and more
+
+## License
+MIT

torch_anatomy-0.1.0/pyproject.toml

@@ -0,0 +1,3 @@
+[build-system]
+requires = ["setuptools", "wheel"]
+build-backend = "setuptools.build_meta"

torch_anatomy-0.1.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

torch_anatomy-0.1.0/setup.py

@@ -0,0 +1,23 @@
+from setuptools import setup, find_packages
+
+setup(
+    name='torch-anatomy',
+    version='0.1.0',
+    description='Layer-by-layer visualizer for PyTorch models',
+    author='Harshal Vilas Kale',
+    packages=find_packages(),
+    install_requires=[
+        'torch',
+        'torchvision',
+        'matplotlib',
+        'numpy',
+        'Pillow',
+        'click'
+    ],
+    entry_points={
+        'console_scripts': [
+            'torch-anatomy=torch_anatomy.cli:main'
+        ]
+    },
+    license="MIT",
+)

torch_anatomy-0.1.0/tests/test_visualizer.py

@@ -0,0 +1,2 @@
+def test_placeholder():
+    assert True 

torch_anatomy-0.1.0/torch_anatomy/__init__.py

@@ -0,0 +1 @@
+from .visualizer import visualize_layers 

torch_anatomy-0.1.0/torch_anatomy/cli.py

@@ -0,0 +1,20 @@
+import click
+import torch
+from torchvision import models
+from .visualizer import visualize_layers
+
+def get_model(model_name):
+    # Simple model loader for demo (expand as needed)
+    if model_name == 'resnet18':
+        return models.resnet18(pretrained=True)
+    raise ValueError(f"Unknown model: {model_name}")
+
+@click.command()
+@click.option('--model', required=True, help='Model name, e.g. resnet18')
+@click.option('--image', required=True, help='Path to input image')
+def main(model, image):
+    model_obj = get_model(model)
+    visualize_layers(model=model_obj, input_image=image)
+
+if __name__ == '__main__':
+    main() 

torch_anatomy-0.1.0/torch_anatomy/utils.py

@@ -0,0 +1,23 @@
+from PIL import Image
+import numpy as np
+import torch
+import torchvision.transforms as T
+
+def load_image(img, size=224):
+    """
+    Loads and preprocesses an image for a PyTorch model.
+    Accepts file path, PIL.Image, or np.ndarray.
+    Returns a torch.Tensor of shape (1, C, H, W)
+    """
+    if isinstance(img, str):
+        img = Image.open(img).convert('RGB')
+    elif isinstance(img, np.ndarray):
+        img = Image.fromarray(img)
+    # Resize and normalize
+    transform = T.Compose([
+        T.Resize((size, size)),
+        T.ToTensor(),
+        T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+    ])
+    tensor = transform(img).unsqueeze(0)
+    return tensor 

torch_anatomy-0.1.0/torch_anatomy/visualizer.py

@@ -0,0 +1,129 @@
+import torch
+import torchvision
+import matplotlib
+matplotlib.use('Agg')  # Set backend to non-interactive
+import matplotlib.pyplot as plt
+import numpy as np
+from PIL import Image
+import os
+from .utils import load_image
+
+def get_activation(name):
+    """Hook to get activations from a layer"""
+    def hook(model, input, output):
+        activations[name] = output.detach()
+    return hook
+
+def visualize_layers(
+    model,
+    input_image,
+    layers_to_show=None,
+    save_dir=None,
+    show=True,
+    channels_per_layer=4,
+    colormap='viridis',
+    show_colorbar=False
+):
+    """
+    Visualize intermediate activations of a PyTorch CNN model layer-by-layer.
+    Args:
+        model: PyTorch model (nn.Module)
+        input_image: Path to image or PIL.Image or np.ndarray
+        layers_to_show: List of layer names/types to visualize (default: Conv, ReLU, Pool)
+        save_dir: If provided, saves images to this directory
+        show: If True, displays the plots
+        channels_per_layer: Number of channels to show per layer (default: 4)
+        colormap: Matplotlib colormap to use (default: 'viridis')
+        show_colorbar: Whether to show colorbar (default: False)
+    """
+    global activations
+    activations = {}
+    
+    # Default layers to show if none specified
+    if layers_to_show is None:
+        layers_to_show = ['Conv2d', 'ReLU', 'MaxPool2d']
+    
+    # Register hooks for all layers
+    hooks = []
+    for name, layer in model.named_modules():
+        if any(layer_type in str(type(layer)) for layer_type in layers_to_show):
+            hooks.append(layer.register_forward_hook(get_activation(name)))
+    
+    # Load and preprocess image
+    input_tensor = load_image(input_image)
+    
+    # Forward pass
+    model.eval()
+    with torch.no_grad():
+        output = model(input_tensor)
+    
+    # Remove hooks
+    for hook in hooks:
+        hook.remove()
+    
+    # Create visualization
+    n_layers = len(activations)
+    if n_layers == 0:
+        print("No matching layers found!")
+        return
+    
+    # Prepare input image for display
+    if isinstance(input_image, str):
+        img_disp = Image.open(input_image).convert('RGB')
+    elif isinstance(input_image, np.ndarray):
+        img_disp = Image.fromarray(input_image)
+    else:
+        img_disp = input_image
+    
+    # Calculate grid size
+    n_cols = channels_per_layer
+    n_rows = n_layers + 1  # +1 for input image
+    
+    plt.figure(figsize=(4*n_cols, 4*n_rows))
+    
+    # Plot input image
+    plt.subplot(n_rows, n_cols, 1)
+    plt.imshow(img_disp)
+    plt.title('Input Image')
+    plt.axis('off')
+    if show_colorbar:
+        plt.colorbar()
+    # Fill rest of first row with blanks if channels_per_layer > 1
+    for i in range(2, n_cols+1):
+        plt.subplot(n_rows, n_cols, i)
+        plt.axis('off')
+    
+    # Plot each layer's activations (top N channels)
+    for row_idx, (name, activation) in enumerate(activations.items(), 1):
+        n_ch = activation.shape[1]
+        for ch in range(min(channels_per_layer, n_ch)):
+            plt.subplot(n_rows, n_cols, row_idx*n_cols + ch + 1)
+            act = activation[0, ch].cpu().numpy()
+            act = (act - act.min()) / (act.max() - act.min() + 1e-8)
+            im = plt.imshow(act, cmap=colormap)
+            plt.title(f'{name}\nChannel {ch} | Shape: {activation.shape}')
+            plt.axis('off')
+            if show_colorbar:
+                plt.colorbar(im, fraction=0.046, pad=0.04)
+        # If channels_per_layer > n_ch, fill rest with blanks
+        for ch in range(n_ch, channels_per_layer):
+            plt.subplot(n_rows, n_cols, row_idx*n_cols + ch + 1)
+            plt.axis('off')
+    
+    plt.tight_layout()
+    
+    # Save if directory provided
+    if save_dir:
+        os.makedirs(save_dir, exist_ok=True)
+        plt.savefig(os.path.join(save_dir, 'layer_visualizations.png'))
+        print(f"Visualizations saved to {save_dir}/layer_visualizations.png")
+    
+    # Show plot
+    if show:
+        try:
+            plt.show()
+        except Exception as e:
+            print(f"Could not display plot: {e}")
+            print("But the visualization has been saved to the output directory!")
+    else:
+        plt.close() 

torch_anatomy-0.1.0/torch_anatomy.egg-info/PKG-INFO

@@ -0,0 +1,18 @@
+Metadata-Version: 2.4
+Name: torch-anatomy
+Version: 0.1.0
+Summary: Layer-by-layer visualizer for PyTorch models
+Author: Harshal Vilas Kale
+License: MIT
+License-File: LICENSE
+Requires-Dist: torch
+Requires-Dist: torchvision
+Requires-Dist: matplotlib
+Requires-Dist: numpy
+Requires-Dist: Pillow
+Requires-Dist: click
+Dynamic: author
+Dynamic: license
+Dynamic: license-file
+Dynamic: requires-dist
+Dynamic: summary

torch_anatomy-0.1.0/torch_anatomy.egg-info/SOURCES.txt

@@ -0,0 +1,15 @@
+LICENSE
+README.md
+pyproject.toml
+setup.py
+tests/test_visualizer.py
+torch_anatomy/__init__.py
+torch_anatomy/cli.py
+torch_anatomy/utils.py
+torch_anatomy/visualizer.py
+torch_anatomy.egg-info/PKG-INFO
+torch_anatomy.egg-info/SOURCES.txt
+torch_anatomy.egg-info/dependency_links.txt
+torch_anatomy.egg-info/entry_points.txt
+torch_anatomy.egg-info/requires.txt
+torch_anatomy.egg-info/top_level.txt

torch_anatomy-0.1.0/torch_anatomy.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

torch_anatomy-0.1.0/torch_anatomy.egg-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+torch-anatomy = torch_anatomy.cli:main

torch_anatomy-0.1.0/torch_anatomy.egg-info/requires.txt

@@ -0,0 +1,6 @@
+torch
+torchvision
+matplotlib
+numpy
+Pillow
+click

torch_anatomy-0.1.0/torch_anatomy.egg-info/top_level.txt

@@ -0,0 +1 @@
+torch_anatomy