core

Some utility functions for working with PyTorch

set_seed

 set_seed (seed:int, deterministic:bool=False)

*Sets the seed for generating random numbers in PyTorch, NumPy, and Python’s random module.

This function is used for reproducibility in stochastic operations, e.g. shuffling in data loaders, random initializations in neural networks, etc.

Note: The deterministic flag does not guarantee complete reproducibility. Operations which rely on CUDA might still produce non-deterministic results.*

	Type	Default	Details
seed	int		The seed value to be set for all random number generators.
deterministic	bool	False	If True, uses deterministic algorithms in PyTorch where possible for reproducibility, at the cost of performance.
Returns	None

seed = 1234
set_seed(seed)

source

pil_to_tensor

 pil_to_tensor (img:<module'PIL.Image'from'/opt/hostedtoolcache/Python/3.1
                0.15/x64/lib/python3.10/site-packages/PIL/Image.py'>,
                mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])

*Converts a PIL image to a normalized and batched PyTorch tensor.

Returns: The normalized and batched tensor.*

	Type	Default	Details
img	PIL.Image		The input PIL image.
mean	list	[0.485, 0.456, 0.406]	The mean values for normalization.
std	list	[0.229, 0.224, 0.225]	The standard deviation values for normalization.

Set the path for the images directory

img_dir = Path('../images/')
img_dir

Path('../images')

Open sample image

img_path = '../images/cat.jpg'
src_img = Image.open(img_path).convert('RGB')
print(f"Image Size: {src_img.size}")
src_img

Image Size: (768, 512)

Convert image to tensor

img_tensor = pil_to_tensor(src_img, [0.5], [0.5])
img_tensor.shape, img_tensor.min(), img_tensor.max()

(torch.Size([1, 3, 512, 768]), tensor(-1.), tensor(1.))

source

tensor_to_pil

 tensor_to_pil (tensor:torch.Tensor)

*Convert a tensor to a PIL image.

Returns: img (PIL.Image): The PIL image*

	Type	Details
tensor	Tensor	The tensor to be converted

Convert tensor to image

tensor_img = tensor_to_pil(transforms.ToTensor()(src_img))
tensor_img

source

iterate_modules

 iterate_modules (module:torch.nn.modules.module.Module)

A generator function that yields the children and grandchildren of a PyTorch module.

	Type	Details
module	Module	A PyTorch module that contains child modules to be iterated over.

from torchvision import models

vgg = models.vgg16(weights=models.VGG16_Weights.IMAGENET1K_V1).features

for index, module in enumerate(iterate_modules(vgg)):
    if type(module) == torch.nn.modules.activation.ReLU:
        print(f"{index}: {module}")

1: ReLU(inplace=True)
3: ReLU(inplace=True)
6: ReLU(inplace=True)
8: ReLU(inplace=True)
11: ReLU(inplace=True)
13: ReLU(inplace=True)
15: ReLU(inplace=True)
18: ReLU(inplace=True)
20: ReLU(inplace=True)
22: ReLU(inplace=True)
25: ReLU(inplace=True)
27: ReLU(inplace=True)
29: ReLU(inplace=True)

Downloading: "https://download.pytorch.org/models/vgg16-397923af.pth" to /home/runner/.cache/torch/hub/checkpoints/vgg16-397923af.pth
  0%|          | 0.00/528M [00:00<?, ?B/s]  3%|2         | 13.8M/528M [00:00<00:03, 143MB/s]  6%|5         | 29.6M/528M [00:00<00:03, 157MB/s] 12%|#1        | 62.8M/528M [00:00<00:02, 244MB/s] 19%|#8        | 97.9M/528M [00:00<00:01, 293MB/s] 25%|##5       | 133M/528M [00:00<00:01, 318MB/s]  32%|###1      | 167M/528M [00:00<00:01, 331MB/s] 38%|###8      | 201M/528M [00:00<00:01, 342MB/s] 45%|####4     | 235M/528M [00:00<00:00, 345MB/s] 51%|#####     | 269M/528M [00:00<00:00, 347MB/s] 57%|#####7    | 303M/528M [00:01<00:00, 350MB/s] 64%|######4   | 338M/528M [00:01<00:00, 355MB/s] 71%|#######   | 372M/528M [00:01<00:00, 356MB/s] 77%|#######6  | 406M/528M [00:01<00:00, 357MB/s] 84%|########3 | 441M/528M [00:01<00:00, 359MB/s] 90%|######### | 475M/528M [00:01<00:00, 358MB/s] 97%|#########6| 509M/528M [00:01<00:00, 357MB/s]100%|##########| 528M/528M [00:01<00:00, 334MB/s]

source

tensor_stats_df

 tensor_stats_df (tensor, attrs=['mean', 'std', 'min', 'max'], shape=True)

Calculate and return statistics of a given tensor as a pandas DataFrame.

	Type	Default	Details
tensor			Input tensor for which statistics are to be calculated.
attrs	list	[‘mean’, ‘std’, ‘min’, ‘max’]	List of statistics to be calculated.
shape	bool	True	If True, include shape of the tensor in the output.

tensor_stats_df(torch.randn(1, 3, 256, 256))

	0
mean	0.003342
std	0.99868
min	-4.558271
max	4.815985
shape	(1, 3, 256, 256)

source

get_torch_device

 get_torch_device ()

*This function returns the device to be used for PyTorch computations.

Returns: str: “mps” if Metal Performance Shaders (MPS) for MacOS is available, “cuda” if CUDA is available, “cpu” otherwise*

get_torch_device()

'cuda'

source

denorm_img_tensor

 denorm_img_tensor (img_tensor:torch.Tensor, mean:list, std:list)

*Denormalize an image tensor.

Returns: torch.Tensor: The tensor representing the denormalized image.*

	Type	Details
img_tensor	Tensor	The tensor representing the normalized image.
mean	list	The mean values used for normalization.
std	list	The standard deviation values used for normalization.

tensor_to_pil(img_tensor)

tensor_to_pil(denorm_img_tensor(img_tensor, [0.5], [0.5]))

source

move_data_to_device

 move_data_to_device (data, device:torch.device)

*Recursively move data to the specified device.

This function takes a data structure (could be a tensor, list, tuple, or dictionary) and moves all tensors within the structure to the given PyTorch device.*

	Type	Details
data		Data to move to the device.
device	device	The PyTorch device to move the data to.

data_tuple = (
    {
        'masks': torch.tensor([[[0, 1, 2], [3, 4, 5], [6, 7, 8]]], dtype=torch.int32),
        'boxes': torch.tensor([[19.9176, 246.6063, 512.0000, 438.8571]]),
        'labels': torch.tensor([0.]),
    },
    {
        'masks': torch.tensor([[[0, 0, 0], [0, 0, 0], [0, 0, 0]]], dtype=torch.int32),
        'boxes': torch.tensor([[19.9176, 246.6063, 512.0000, 438.8571]]),
        'labels': torch.tensor([0.]),
    },
)

move_data_to_device(data_tuple, 'cpu')

({'masks': tensor([[[0, 1, 2],
           [3, 4, 5],
           [6, 7, 8]]], dtype=torch.int32),
  'boxes': tensor([[ 19.9176, 246.6063, 512.0000, 438.8571]]),
  'labels': tensor([0.])},
 {'masks': tensor([[[0, 0, 0],
           [0, 0, 0],
           [0, 0, 0]]], dtype=torch.int32),
  'boxes': tensor([[ 19.9176, 246.6063, 512.0000, 438.8571]]),
  'labels': tensor([0.])})

source

ImageDataset

 ImageDataset (image_paths, transform=None)

A PyTorch Dataset for RGB images.

source

compute_mean_std

 compute_mean_std (image_paths:List[pathlib.Path], batch_size:int=32,
                   num_workers:int=0, image_size:int=224, transform:torchv
                   ision.transforms.transforms.Compose=None)

Computes the mean and standard deviation of images provided in image_paths.

	Type	Default	Details
image_paths	List		List of image file paths.
batch_size	int	32	Number of images to process in a batch.
num_workers	int	0	Number of subprocesses to use for data loading.
image_size	int	224	Size to resize images to.
transform	Compose	None	Torchvision transforms to apply to the images.
Returns	dict		Dictionary containing ‘mean’ and ‘std’ values.

compute_mean_std(img_paths, batch_size=64)