showtens 0.0__py3-none-any.whl

showtens/__init__.py

@@ -0,0 +1,3 @@
+from .util import gridify
+from .images import showImage, saveImage
+from .videos import saveVideo

showtens/images.py

@@ -0,0 +1,115 @@
+import os
+import matplotlib.pyplot as plt
+from .imports import import_torch, import_torchvision
+from .util import gridify, _create_folder
+
+torch = import_torch()
+torchvision = import_torchvision()
+
+
+@torch.no_grad()
+def showImage(tensor, columns=None, colorbar=False, max_width=None, padding=3, pad_value=0.0):
+    """ "
+    Shows tensor as an image using pyplot.
+    Any extra dimensions (*,C,H,W) are treated as batch dimensions.
+
+    Args:
+    tensor : (H,W) or (C,H,W) or (*,C,H,W) tensor to display
+    columns : number of columns to use for the grid of images (default 8 or less)
+    colorbar : whether to add a colorbar to the image, only works for grayscale images (default False)
+    max_width : maximum width of the image
+    padding : number of pixels between images in the grid
+    pad_value : inter-padding value for the grid of images
+    """
+    tensor = _format_image(
+        tensor, columns=columns, max_width=max_width, padding=padding, pad_value=pad_value
+    )  # (C,H',W') ready to show
+
+    plt.imshow(tensor.permute((1, 2, 0)))
+    plt.axis("off")
+    if tensor.shape[0] == 1 and colorbar:
+        plt.colorbar()
+    plt.show()
+
+
+@torch.no_grad()
+def saveImage(
+    tensor,
+    folder,
+    name="imagetensor",
+    columns=None,
+    colorbar=False,
+    max_width=None,
+    padding=3,
+    pad_value=0.0,
+    create_folder=True,
+):
+    """
+    Saves tensor as a png image using pyplot.
+    Any extra dimensions (*,C,H,W) are treated as batch dimensions.
+
+    Args:
+    tensor : (H,W) or (C,H,W) or (*,C,H,W) tensor to display
+    folder : relative path of folder where to save the image
+    name : name of the image (do not include extension)
+    columns : number of columns to use for the grid of images (default 8 or less)
+    colorbar : whether to add a colorbar to the image, only works for grayscale images (default False)
+    max_width : maximum width of the image
+    padding : number of pixels between images in the grid
+    pad_value : inter-padding value for the grid of images
+    create_folder : whether to create the folder if it does not exist (default True)
+    """
+    _create_folder(folder, create_folder)
+
+    tensor = _format_image(
+        tensor, columns=columns, max_width=max_width, padding=padding, pad_value=pad_value
+    )  # (C,H',W') ready to save
+    plt.imshow(tensor.permute((1, 2, 0)))
+    plt.axis("off")
+    if tensor.shape[0] == 1 and colorbar:
+        plt.colorbar()
+
+    plt.savefig(os.path.join(folder, f"{name}.png"), bbox_inches="tight", pad_inches=0)
+
+
+@torch.no_grad()
+def _format_image(tensor, columns=None, max_width=None, padding=3, pad_value=0.0):
+    """ "
+    Shows tensor as an image using pyplot.
+    Any extra dimensions (*,C,H,W) are treated as batch dimensions.
+
+    Args:
+    tensor : (H,W) or (C,H,W) or (*,C,H,W) tensor to display
+    columns : number of columns to use for the grid of images (default 8 or less)
+    max_width : maximum width of the image
+    padding : number of pixels between images in the grid
+    pad_value : inter-padding value for the grid of images
+    """
+    tensor = tensor.detach().cpu()
+
+    extra_params = dict(columns=columns, max_width=max_width, pad_value=pad_value, padding=padding)
+    if len(tensor.shape) == 2:
+        # Add batch and channel dimensions
+        return _format_image(tensor[None, :, :], **extra_params)
+    elif len(tensor.shape) == 3:
+        # Reached (C,H,W)
+        return tensor
+    elif len(tensor.shape) == 4:
+        # Gridify assuming (B,C,H,W)
+        B = tensor.shape[0]
+        if columns is not None:
+            numCol = columns
+        else:
+            numCol = min(8, B)
+        tensor = gridify(tensor, columns=numCol, max_width=max_width, pad_value=pad_value, padding=padding)
+
+        return _format_image(tensor, **extra_params)
+    elif len(tensor.shape) > 4:
+        # Collapse extra dimension to batch
+        tensor = tensor.reshape(
+            (-1, tensor.shape[-3], tensor.shape[-2], tensor.shape[-1])
+        )  # assume all batch dimensions
+        print("Assuming extra dimension are all batch dimensions, newshape : ", tensor.shape)
+        return _format_image(tensor, **extra_params)
+    else:
+        raise Exception(f"Tensor shape should be (H,W), (C,H,W) or (*,C,H,W), but got : {tensor.shape} !")

showtens/imports.py

@@ -0,0 +1,35 @@
+def import_torch():
+    """
+    Lazily import torch, giving helpful error message if not installed.
+    """
+    try:
+        import torch
+
+        return torch
+    except ImportError:
+        raise ImportError(
+            "PyTorch is not installed. ShowTens requires PyTorch. "
+            "Please install it with one of the following commands:\n\n"
+            "- For CPU only: pip install torch\n"
+            "- For CUDA support: Visit https://pytorch.org for installation instructions "
+            "specific to your system and CUDA version."
+        )
+
+
+def import_torchvision():
+    """
+    Lazily import torchvision.transforms, giving helpful error message if not installed.
+    """
+    try:
+        import torchvision
+
+        return torchvision
+    except ImportError:
+        raise ImportError(
+            "torchvision is not installed. ShowTens requires torchvision. "
+            "Please install it with the following command:\n\n"
+            "pip install torchvision"
+        )
+
+
+# torch = import_torch()

showtens/util.py

@@ -0,0 +1,72 @@
+from .imports import import_torch, import_torchvision
+import os
+
+torch = import_torch()
+
+
+@torch.no_grad()
+def gridify(tensor: torch.Tensor, max_width=None, columns=None, padding=3, pad_value=0.0):
+    """
+    Makes a grid of images/videos from a batch of images.
+    Like torchvision's make_grid, but more flexible.
+    Accepts (B,*,H,W)
+
+    Args:
+        tensor : (B,*,H,W) tensor
+        max_width : max width of the output grid. Resizes images to fit the width
+        columns : number of columns of the grid. If None, uses 8 or less
+        padding : padding to add to the images
+        pad_value : color of the padding
+
+    Returns:
+        (*,H',W') tensor, representing the grid of images/videos
+    """
+    transf = import_torchvision().transforms
+
+    B, H, W = tensor.shape[0], tensor.shape[-2], tensor.shape[-1]
+    device = tensor.device
+    if columns is not None:
+        numCol = columns
+    else:
+        numCol = min(8, B)
+
+    black_cols = (-B) % numCol
+    tensor = torch.cat(
+        [tensor, torch.zeros(black_cols, *tensor.shape[1:], device=device)], dim=0
+    )  # (B',*,H,W)
+    tensor = transf.Pad(padding, fill=pad_value)(tensor)  # (B',*,H+padding*2,W+padding*2)
+
+    B, H, W = tensor.shape[0], tensor.shape[-2], tensor.shape[-1]
+    rest_dim = tensor.shape[1:-2]
+
+    rest_dim_prod = 1
+    for dim in rest_dim:
+        rest_dim_prod *= dim
+
+    if max_width is not None:
+        resize_ratio = max_width / (W * numCol)
+        if resize_ratio < 1:
+            indiv_tens_size = int(H * resize_ratio), int(W * resize_ratio)
+            tensor = tensor.reshape((B, rest_dim_prod, H, W))
+            tensor = transf.Resize(indiv_tens_size, antialias=True)(tensor)  # (B',rest_dim_prod,H',W')
+
+    B, H, W = tensor.shape[0], tensor.shape[-2], tensor.shape[-1]
+    assert B % numCol == 0
+
+    numRows = B // numCol
+
+    tensor = tensor.reshape((numRows, numCol, rest_dim_prod, H, W))  # (numRows,numCol,rest_dim_prod,H',W')
+    tensor = torch.einsum("nmrhw->rnhmw", tensor)  # (rest_prod,numRows,H',numCol,W')
+    tensor = tensor.reshape((rest_dim_prod, numRows * H, numCol * W))  # (rest_prod,numRows*H,numCol*W)
+    tensor = tensor.reshape((*rest_dim, numRows * H, numCol * W))  # (*,numRows*H,numCol*W)
+
+    return tensor
+
+
+@torch.no_grad()
+def _create_folder(folder: str, create_folder: bool = True):
+    if create_folder:
+        os.makedirs(folder, exist_ok=True)
+    else:
+        if not (os.path.exists(folder)):
+            raise FileNotFoundError(f"Folder {folder} does not exist !")

showtens/videos.py

@@ -0,0 +1,104 @@
+import os
+from .util import gridify, import_torch, _create_folder
+import cv2, numpy as np
+
+torch = import_torch()
+
+
+@torch.no_grad()
+def showVideo(tensor, fps=30, columns=None, max_width=None, padding=3, pad_value=0.0):
+    """
+    Shows tensor as a video. Accepts both (T,H,W), (T,3,H,W) and (*,T,3,H,W) float tensors.
+
+    Args:
+    tensor : (T,H,W) or (T,3,H,W) or (*,T,3,H,W) float tensor
+    columns : number of columns to use for the grid of videos (default 8 or less)
+    fps : fps of the video (default 30)
+    out_size : Height of output video (height adapts to not deform videos) (default 800)
+    """
+    return NotImplementedError("showVideo not implemented yet, use saveVideo instead")
+
+
+@torch.no_grad()
+def saveVideo(
+    tensor,
+    folder,
+    name="videotensor",
+    fps=30,
+    columns=None,
+    max_width=None,
+    padding=3,
+    pad_value=0.0,
+    create_folder=True,
+):
+    """
+    Saves tensor as a video. Accepts both (T,H,W), (T,3,H,W) and (*,T,3,H,W) float tensors.
+    Assumes that the tensor value are in [0,1], clips them otherwise.
+
+    Args:
+    tensor : (T,H,W) or (T,3,H,W) or (*,T,3,H,W) float tensor
+    folder : path to save the video
+    name : name of the video
+    fps : fps of the video (default 30)
+    columns : number of columns to use for the grid of videos (default 8 or less)
+    max_width : maximum width of the image
+    padding : number of pixels between images in the grid
+    pad_value : inter-padding value for the grid of images
+    """
+    _create_folder(folder, create_folder)
+
+    tensor = _format_video(
+        tensor, columns=columns, fps=fps, max_width=max_width, padding=padding, pad_value=pad_value
+    )  # (T,3,H',W') ready to save
+    T, C, H, W = tensor.shape
+    output_file = os.path.join(folder, f"{name}.mp4")
+    fourcc = cv2.VideoWriter_fourcc(*"mp4v")
+    video = cv2.VideoWriter(output_file, fourcc, fps, (W, H))
+
+    to_save = (255 * tensor.permute(0, 2, 3, 1).cpu().numpy()).astype(np.uint8)
+
+    for t in range(T):
+        frame = to_save[t]
+        frame = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)
+        video.write(frame)
+
+    video.release()
+
+
+def _format_video(tensor, columns=None, fps=30, max_width=None, padding=3, pad_value=0.0):
+    """
+    Formats tensor as a video. Accepts both (T,H,W), (T,3,H,W) and (*,T,3,H,W) float tensors.
+    Assumes that the tensor value are in [0,1], clips them otherwise.
+
+    Args:
+    tensor : (T,H,W) or (T,3,H,W) or (*,T,3,H,W) float tensor
+    columns : number of columns to use for the grid of videos (default 8 or less)
+    fps : fps of the video (default 30)
+    max_width : maximum width of the image
+    padding : number of pixels between images in the grid
+    pad_value : inter-padding value for the grid of images
+    """
+    tensor = tensor.detach().cpu()
+    extra_params = dict(columns=columns, fps=fps, max_width=max_width, pad_value=pad_value, padding=padding)
+
+    if len(tensor.shape) == 3:
+        # add channel dimension
+        tensor = tensor[:, None, :, :].expand(-1, 3, -1, -1)  # (T,3,H,W)
+        return _format_video(tensor, **extra_params)
+    elif len(tensor.shape) == 4:
+        if tensor.shape[1] == 1:
+            print("Assuming gray-scale video")
+            tensor = tensor.expand(-1, 3, -1, -1)  # (T,3,H,W)
+        assert tensor.shape[1] == 3, f"Tensor shape should be (T,H,W), (T,3,H,W) or (*,T,3,H,W) !"
+        # A single video
+        return tensor
+    elif len(tensor.shape) == 5:
+        tensor = gridify(tensor, max_width=max_width, columns=columns, padding=padding, pad_value=pad_value)
+        return _format_video(tensor, **extra_params)
+    elif len(tensor.shape) > 5:
+        tensor = tensor.reshape((-1, *tensor.shape[-4:]))
+        return _format_video(tensor, **extra_params)
+    else:
+        raise ValueError(
+            f"Tensor shape should be (T,H,W), (T,3,H,W) or (*,T,3,H,W), but got : {tensor.shape} !"
+        )

showtens-0.0.dist-info/METADATA

@@ -0,0 +1,49 @@
+Metadata-Version: 2.1
+Name: showtens
+Version: 0.0
+Summary: Visualize torch tensors EASILY.
+Author-email: Vassilis Papadopoulos <vassilis.physics@gmail.com>
+License: MIT License
+Project-URL: Homepage, https://github.com/frotaur/showtens
+Project-URL: Bug Tracker, https://github.com/frotaur/showtens/issues
+Classifier: Programming Language :: Python :: 3
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Operating System :: OS Independent
+Classifier: Development Status :: 3 - Alpha
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+Requires-Dist: matplotlib
+Requires-Dist: opencv-python
+Requires-Dist: numpy
+
+# ShowTens : visualize torch tensors EASILY
+
+ShowTens is a simple pytorch package that allows painless and flexible visualizations of image and video tensors.
+
+\<ADD VISUALIZATION VIDEO HERE\>
+
+## Installation
+
+`pip install showtens`
+
+Make sure `torch`and `torchvision` are installed, as the package depends on them.
+
+## Usage
+```python
+import torch
+from showTens import showImage
+
+image1 = torch.rand((3,100,100)) # (C,H,W) image
+showImage(image1) # Displays the image using matplotlib
+image2 = torch.rand((4,4,3,100,100)) # (B1,B2,C,H,W), two batch dimensions
+showImage(image2,colums=4) # Will display as a 4*4 grid
+
+from showTens import saveImage
+saveImage(tensor=image1,folder='saved_images',name='imagetensor')
+
+from showTens import saveVideo
+video1 = torch.rand((60,3,200,200))
+saveVideo(tensor=video1,folder='save_videos',name='videotensor',fps=30)
+video2 = torch.rand((4,3,200,200)) # (B,T,C,H,W), batch of videos
+saveVideo(tensor=video2,folder='save_videos',name='videobatch',fps=30,columns=2) # 2*2 video grid
+```

showtens-0.0.dist-info/RECORD

@@ -0,0 +1,9 @@
+showtens/__init__.py,sha256=r0JRXhXHlVJmITQlPQINI0rtM7vf_Mn9EELcU_Wh4g8,98
+showtens/images.py,sha256=-9RGBhpBXmXjV8d5c-M8df4mUddoZgivbaaCIwf01_A,4380
+showtens/imports.py,sha256=xx9G4Ap3uTAlXxh7LaNvIz-P0rSMsiMxeGQc5oTo3Dw,1056
+showtens/util.py,sha256=g_2B-KV_OSTakn9P6Ku2u6FTkxhbZfhywYpgnch4fgA,2592
+showtens/videos.py,sha256=Gpvisiqq_7w0a9RhkXJOXEf9qdPW3ROTwrV-UYJDyIE,3995
+showtens-0.0.dist-info/METADATA,sha256=6bGZPusjOFlZcMCq77U1bb1Y-Gdy8iicehfUbQLCo7E,1708
+showtens-0.0.dist-info/WHEEL,sha256=P9jw-gEje8ByB7_hXoICnHtVCrEwMQh-630tKvQWehc,91
+showtens-0.0.dist-info/top_level.txt,sha256=1SACMFm7OZu9Dl-DE-HZyHym7Lan02cXwMWAMWj5zsA,9
+showtens-0.0.dist-info/RECORD,,

showtens-0.0.dist-info/WHEEL

@@ -0,0 +1,5 @@
+Wheel-Version: 1.0
+Generator: setuptools (75.3.0)
+Root-Is-Purelib: true
+Tag: py3-none-any
+

showtens-0.0.dist-info/top_level.txt

@@ -0,0 +1 @@
+showtens