utils.py 21.9 KB
Newer Older
1
import math
2
import pathlib
3
import warnings
Kai Zhang's avatar
Kai Zhang committed
4
from types import FunctionType
5
from typing import Any, BinaryIO, List, Optional, Tuple, Union
6

7
import numpy as np
8
import torch
9
from PIL import Image, ImageColor, ImageDraw, ImageFont
10

11
12
13
14
15
16
17
18
__all__ = [
    "make_grid",
    "save_image",
    "draw_bounding_boxes",
    "draw_segmentation_masks",
    "draw_keypoints",
    "flow_to_image",
]
19

20

21
@torch.no_grad()
22
def make_grid(
23
    tensor: Union[torch.Tensor, List[torch.Tensor]],
24
25
26
    nrow: int = 8,
    padding: int = 2,
    normalize: bool = False,
27
    value_range: Optional[Tuple[int, int]] = None,
28
    scale_each: bool = False,
29
    pad_value: float = 0.0,
30
    **kwargs,
31
) -> torch.Tensor:
32
33
    """
    Make a grid of images.
34

35
36
37
    Args:
        tensor (Tensor or list): 4D mini-batch Tensor of shape (B x C x H x W)
            or a list of images all of the same size.
38
        nrow (int, optional): Number of images displayed in each row of the grid.
Tongzhou Wang's avatar
Tongzhou Wang committed
39
40
            The final grid size is ``(B / nrow, nrow)``. Default: ``8``.
        padding (int, optional): amount of padding. Default: ``2``.
41
        normalize (bool, optional): If True, shift the image to the range (0, 1),
42
            by the min and max values specified by ``value_range``. Default: ``False``.
43
        value_range (tuple, optional): tuple (min, max) where min and max are numbers,
44
45
            then these numbers are used to normalize the image. By default, min and max
            are computed from the tensor.
46
47
48
49
        range (tuple. optional):
            .. warning::
                This parameter was deprecated in ``0.12`` and will be removed in ``0.14``. Please use ``value_range``
                instead.
Tongzhou Wang's avatar
Tongzhou Wang committed
50
51
52
        scale_each (bool, optional): If ``True``, scale each image in the batch of
            images separately rather than the (min, max) over all images. Default: ``False``.
        pad_value (float, optional): Value for the padded pixels. Default: ``0``.
53

54
55
    Returns:
        grid (Tensor): the tensor containing grid of images.
56
    """
Kai Zhang's avatar
Kai Zhang committed
57
58
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(make_grid)
59
60
61
62
63
64
65
    if not torch.is_tensor(tensor):
        if isinstance(tensor, list):
            for t in tensor:
                if not torch.is_tensor(t):
                    raise TypeError(f"tensor or list of tensors expected, got a list containing {type(t)}")
        else:
            raise TypeError(f"tensor or list of tensors expected, got {type(tensor)}")
66
67

    if "range" in kwargs.keys():
68
69
70
71
        warnings.warn(
            "The parameter 'range' is deprecated since 0.12 and will be removed in 0.14. "
            "Please use 'value_range' instead."
        )
72
        value_range = kwargs["range"]
73

74
    # if list of tensors, convert to a 4D mini-batch Tensor
75
    if isinstance(tensor, list):
76
        tensor = torch.stack(tensor, dim=0)
77

78
    if tensor.dim() == 2:  # single image H x W
79
        tensor = tensor.unsqueeze(0)
80
    if tensor.dim() == 3:  # single image
81
        if tensor.size(0) == 1:  # if single-channel, convert to 3-channel
Adam Lerer's avatar
Adam Lerer committed
82
            tensor = torch.cat((tensor, tensor, tensor), 0)
83
        tensor = tensor.unsqueeze(0)
84

85
    if tensor.dim() == 4 and tensor.size(1) == 1:  # single-channel images
86
        tensor = torch.cat((tensor, tensor, tensor), 1)
87
88

    if normalize is True:
89
        tensor = tensor.clone()  # avoid modifying tensor in-place
90
91
        if value_range is not None and not isinstance(value_range, tuple):
            raise TypeError("value_range has to be a tuple (min, max) if specified. min and max are numbers")
92

93
94
95
        def norm_ip(img, low, high):
            img.clamp_(min=low, max=high)
            img.sub_(low).div_(max(high - low, 1e-5))
96

97
98
99
        def norm_range(t, value_range):
            if value_range is not None:
                norm_ip(t, value_range[0], value_range[1])
100
            else:
101
                norm_ip(t, float(t.min()), float(t.max()))
102
103
104

        if scale_each is True:
            for t in tensor:  # loop over mini-batch dimension
105
                norm_range(t, value_range)
106
        else:
107
            norm_range(tensor, value_range)
108

109
110
    if not isinstance(tensor, torch.Tensor):
        raise TypeError("tensor should be of type torch.Tensor")
111
    if tensor.size(0) == 1:
112
        return tensor.squeeze(0)
113

114
115
116
    # make the mini-batch of images into a grid
    nmaps = tensor.size(0)
    xmaps = min(nrow, nmaps)
117
    ymaps = int(math.ceil(float(nmaps) / xmaps))
118
    height, width = int(tensor.size(2) + padding), int(tensor.size(3) + padding)
119
120
    num_channels = tensor.size(1)
    grid = tensor.new_full((num_channels, height * ymaps + padding, width * xmaps + padding), pad_value)
121
    k = 0
122
123
    for y in range(ymaps):
        for x in range(xmaps):
124
125
            if k >= nmaps:
                break
126
127
128
129
130
            # Tensor.copy_() is a valid method but seems to be missing from the stubs
            # https://pytorch.org/docs/stable/tensors.html#torch.Tensor.copy_
            grid.narrow(1, y * height + padding, height - padding).narrow(  # type: ignore[attr-defined]
                2, x * width + padding, width - padding
            ).copy_(tensor[k])
131
132
133
134
            k = k + 1
    return grid


135
@torch.no_grad()
136
def save_image(
137
    tensor: Union[torch.Tensor, List[torch.Tensor]],
138
    fp: Union[str, pathlib.Path, BinaryIO],
139
    format: Optional[str] = None,
140
    **kwargs,
141
) -> None:
142
143
    """
    Save a given Tensor into an image file.
144
145
146
147

    Args:
        tensor (Tensor or list): Image to be saved. If given a mini-batch tensor,
            saves the tensor as a grid of images by calling ``make_grid``.
148
        fp (string or file object): A filename or a file object
149
150
        format(Optional):  If omitted, the format to use is determined from the filename extension.
            If a file object was used instead of a filename, this parameter should always be used.
151
        **kwargs: Other arguments are documented in ``make_grid``.
152
    """
153

Kai Zhang's avatar
Kai Zhang committed
154
155
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(save_image)
156
    grid = make_grid(tensor, **kwargs)
157
    # Add 0.5 after unnormalizing to [0, 255] to round to nearest integer
158
    ndarr = grid.mul(255).add_(0.5).clamp_(0, 255).permute(1, 2, 0).to("cpu", torch.uint8).numpy()
159
    im = Image.fromarray(ndarr)
160
    im.save(fp, format=format)
161
162
163
164
165
166
167


@torch.no_grad()
def draw_bounding_boxes(
    image: torch.Tensor,
    boxes: torch.Tensor,
    labels: Optional[List[str]] = None,
168
    colors: Optional[Union[List[Union[str, Tuple[int, int, int]]], str, Tuple[int, int, int]]] = None,
169
    fill: Optional[bool] = False,
170
171
    width: int = 1,
    font: Optional[str] = None,
172
    font_size: Optional[int] = None,
173
174
175
176
177
) -> torch.Tensor:

    """
    Draws bounding boxes on given image.
    The values of the input image should be uint8 between 0 and 255.
178
    If fill is True, Resulting Tensor should be saved as PNG image.
179
180

    Args:
181
        image (Tensor): Tensor of shape (C x H x W) and dtype uint8.
182
        boxes (Tensor): Tensor of size (N, 4) containing bounding boxes in (xmin, ymin, xmax, ymax) format. Note that
183
184
185
            the boxes are absolute coordinates with respect to the image. In other words: `0 <= xmin < xmax < W` and
            `0 <= ymin < ymax < H`.
        labels (List[str]): List containing the labels of bounding boxes.
186
187
188
        colors (color or list of colors, optional): List containing the colors
            of the boxes or single color for all boxes. The color can be represented as
            PIL strings e.g. "red" or "#FF00FF", or as RGB tuples e.g. ``(240, 10, 157)``.
189
            By default, random colors are generated for boxes.
190
        fill (bool): If `True` fills the bounding box with specified color.
191
192
193
194
195
        width (int): Width of bounding box.
        font (str): A filename containing a TrueType font. If the file is not found in this filename, the loader may
            also search in other directories, such as the `fonts/` directory on Windows or `/Library/Fonts/`,
            `/System/Library/Fonts/` and `~/Library/Fonts/` on macOS.
        font_size (int): The requested font size in points.
196
197
198

    Returns:
        img (Tensor[C, H, W]): Image Tensor of dtype uint8 with bounding boxes plotted.
199
200
    """

Kai Zhang's avatar
Kai Zhang committed
201
202
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(draw_bounding_boxes)
203
204
205
206
207
208
    if not isinstance(image, torch.Tensor):
        raise TypeError(f"Tensor expected, got {type(image)}")
    elif image.dtype != torch.uint8:
        raise ValueError(f"Tensor uint8 expected, got {image.dtype}")
    elif image.dim() != 3:
        raise ValueError("Pass individual images, not batches")
209
210
    elif image.size(0) not in {1, 3}:
        raise ValueError("Only grayscale and RGB images are supported")
211
212
213
214
    elif (boxes[:, 0] > boxes[:, 2]).any() or (boxes[:, 1] > boxes[:, 3]).any():
        raise ValueError(
            "Boxes need to be in (xmin, ymin, xmax, ymax) format. Use torchvision.ops.box_convert to convert them"
        )
215

216
217
    num_boxes = boxes.shape[0]

218
219
220
221
    if num_boxes == 0:
        warnings.warn("boxes doesn't contain any box. No box was drawn")
        return image

222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
    if labels is None:
        labels: Union[List[str], List[None]] = [None] * num_boxes  # type: ignore[no-redef]
    elif len(labels) != num_boxes:
        raise ValueError(
            f"Number of boxes ({num_boxes}) and labels ({len(labels)}) mismatch. Please specify labels for each box."
        )

    if colors is None:
        colors = _generate_color_palette(num_boxes)
    elif isinstance(colors, list):
        if len(colors) < num_boxes:
            raise ValueError(f"Number of colors ({len(colors)}) is less than number of boxes ({num_boxes}). ")
    else:  # colors specifies a single color for all boxes
        colors = [colors] * num_boxes

    colors = [(ImageColor.getrgb(color) if isinstance(color, str) else color) for color in colors]

239
240
241
242
243
244
245
    if font is None:
        if font_size is not None:
            warnings.warn("Argument 'font_size' will be ignored since 'font' is not set.")
        txt_font = ImageFont.load_default()
    else:
        txt_font = ImageFont.truetype(font=font, size=font_size or 10)

246
    # Handle Grayscale images
247
248
    if image.size(0) == 1:
        image = torch.tile(image, (3, 1, 1))
249

250
    ndarr = image.permute(1, 2, 0).cpu().numpy()
251
252
253
    img_to_draw = Image.fromarray(ndarr)
    img_boxes = boxes.to(torch.int64).tolist()

254
255
256
257
258
    if fill:
        draw = ImageDraw.Draw(img_to_draw, "RGBA")
    else:
        draw = ImageDraw.Draw(img_to_draw)

259
    for bbox, color, label in zip(img_boxes, colors, labels):  # type: ignore[arg-type]
260
        if fill:
261
            fill_color = color + (100,)
262
263
264
            draw.rectangle(bbox, width=width, outline=color, fill=fill_color)
        else:
            draw.rectangle(bbox, width=width, outline=color)
265

266
        if label is not None:
267
            margin = width + 1
268
            draw.text((bbox[0] + margin, bbox[1] + margin), label, fill=color, font=txt_font)
269

270
    return torch.from_numpy(np.array(img_to_draw)).permute(2, 0, 1).to(dtype=torch.uint8)
271
272
273
274
275
276


@torch.no_grad()
def draw_segmentation_masks(
    image: torch.Tensor,
    masks: torch.Tensor,
277
    alpha: float = 0.8,
278
    colors: Optional[Union[List[Union[str, Tuple[int, int, int]]], str, Tuple[int, int, int]]] = None,
279
280
281
282
283
284
285
) -> torch.Tensor:

    """
    Draws segmentation masks on given RGB image.
    The values of the input image should be uint8 between 0 and 255.

    Args:
286
287
288
289
        image (Tensor): Tensor of shape (3, H, W) and dtype uint8.
        masks (Tensor): Tensor of shape (num_masks, H, W) or (H, W) and dtype bool.
        alpha (float): Float number between 0 and 1 denoting the transparency of the masks.
            0 means full transparency, 1 means no transparency.
290
291
292
293
        colors (color or list of colors, optional): List containing the colors
            of the masks or single color for all masks. The color can be represented as
            PIL strings e.g. "red" or "#FF00FF", or as RGB tuples e.g. ``(240, 10, 157)``.
            By default, random colors are generated for each mask.
294
295

    Returns:
296
        img (Tensor[C, H, W]): Image Tensor, with segmentation masks drawn on top.
297
298
    """

Kai Zhang's avatar
Kai Zhang committed
299
300
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(draw_segmentation_masks)
301
    if not isinstance(image, torch.Tensor):
302
        raise TypeError(f"The image must be a tensor, got {type(image)}")
303
    elif image.dtype != torch.uint8:
304
        raise ValueError(f"The image dtype must be uint8, got {image.dtype}")
305
306
307
308
    elif image.dim() != 3:
        raise ValueError("Pass individual images, not batches")
    elif image.size()[0] != 3:
        raise ValueError("Pass an RGB image. Other Image formats are not supported")
309
310
311
312
313
314
315
316
    if masks.ndim == 2:
        masks = masks[None, :, :]
    if masks.ndim != 3:
        raise ValueError("masks must be of shape (H, W) or (batch_size, H, W)")
    if masks.dtype != torch.bool:
        raise ValueError(f"The masks must be of dtype bool. Got {masks.dtype}")
    if masks.shape[-2:] != image.shape[-2:]:
        raise ValueError("The image and the masks must have the same height and width")
317
318

    num_masks = masks.size()[0]
319
320
    if colors is not None and num_masks > len(colors):
        raise ValueError(f"There are more masks ({num_masks}) than colors ({len(colors)})")
321

322
323
324
325
    if num_masks == 0:
        warnings.warn("masks doesn't contain any mask. No mask was drawn")
        return image

326
    if colors is None:
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
        colors = _generate_color_palette(num_masks)

    if not isinstance(colors, list):
        colors = [colors]
    if not isinstance(colors[0], (tuple, str)):
        raise ValueError("colors must be a tuple or a string, or a list thereof")
    if isinstance(colors[0], tuple) and len(colors[0]) != 3:
        raise ValueError("It seems that you passed a tuple of colors instead of a list of colors")

    out_dtype = torch.uint8

    colors_ = []
    for color in colors:
        if isinstance(color, str):
            color = ImageColor.getrgb(color)
342
        colors_.append(torch.tensor(color, dtype=out_dtype))
343

344
345
346
347
    img_to_draw = image.detach().clone()
    # TODO: There might be a way to vectorize this
    for mask, color in zip(masks, colors_):
        img_to_draw[:, mask] = color[:, None]
348

349
350
    out = image * (1 - alpha) + img_to_draw * alpha
    return out.to(out_dtype)
351
352


353
354
355
356
@torch.no_grad()
def draw_keypoints(
    image: torch.Tensor,
    keypoints: torch.Tensor,
357
    connectivity: Optional[List[Tuple[int, int]]] = None,
358
359
360
361
362
363
364
365
366
367
368
369
370
    colors: Optional[Union[str, Tuple[int, int, int]]] = None,
    radius: int = 2,
    width: int = 3,
) -> torch.Tensor:

    """
    Draws Keypoints on given RGB image.
    The values of the input image should be uint8 between 0 and 255.

    Args:
        image (Tensor): Tensor of shape (3, H, W) and dtype uint8.
        keypoints (Tensor): Tensor of shape (num_instances, K, 2) the K keypoints location for each of the N instances,
            in the format [x, y].
371
        connectivity (List[Tuple[int, int]]]): A List of tuple where,
372
373
374
375
376
377
378
379
380
381
            each tuple contains pair of keypoints to be connected.
        colors (str, Tuple): The color can be represented as
            PIL strings e.g. "red" or "#FF00FF", or as RGB tuples e.g. ``(240, 10, 157)``.
        radius (int): Integer denoting radius of keypoint.
        width (int): Integer denoting width of line connecting keypoints.

    Returns:
        img (Tensor[C, H, W]): Image Tensor of dtype uint8 with keypoints drawn.
    """

Kai Zhang's avatar
Kai Zhang committed
382
383
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(draw_keypoints)
384
385
386
387
388
389
390
391
392
393
394
395
    if not isinstance(image, torch.Tensor):
        raise TypeError(f"The image must be a tensor, got {type(image)}")
    elif image.dtype != torch.uint8:
        raise ValueError(f"The image dtype must be uint8, got {image.dtype}")
    elif image.dim() != 3:
        raise ValueError("Pass individual images, not batches")
    elif image.size()[0] != 3:
        raise ValueError("Pass an RGB image. Other Image formats are not supported")

    if keypoints.ndim != 3:
        raise ValueError("keypoints must be of shape (num_instances, K, 2)")

396
    ndarr = image.permute(1, 2, 0).cpu().numpy()
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
    img_to_draw = Image.fromarray(ndarr)
    draw = ImageDraw.Draw(img_to_draw)
    img_kpts = keypoints.to(torch.int64).tolist()

    for kpt_id, kpt_inst in enumerate(img_kpts):
        for inst_id, kpt in enumerate(kpt_inst):
            x1 = kpt[0] - radius
            x2 = kpt[0] + radius
            y1 = kpt[1] - radius
            y2 = kpt[1] + radius
            draw.ellipse([x1, y1, x2, y2], fill=colors, outline=None, width=0)

        if connectivity:
            for connection in connectivity:
                start_pt_x = kpt_inst[connection[0]][0]
                start_pt_y = kpt_inst[connection[0]][1]

                end_pt_x = kpt_inst[connection[1]][0]
                end_pt_y = kpt_inst[connection[1]][1]

                draw.line(
                    ((start_pt_x, start_pt_y), (end_pt_x, end_pt_y)),
                    width=width,
                )

    return torch.from_numpy(np.array(img_to_draw)).permute(2, 0, 1).to(dtype=torch.uint8)


425
426
427
428
429
430
431
432
# Flow visualization code adapted from https://github.com/tomrunia/OpticalFlow_Visualization
@torch.no_grad()
def flow_to_image(flow: torch.Tensor) -> torch.Tensor:

    """
    Converts a flow to an RGB image.

    Args:
433
        flow (Tensor): Flow of shape (N, 2, H, W) or (2, H, W) and dtype torch.float.
434
435

    Returns:
436
437
        img (Tensor): Image Tensor of dtype uint8 where each color corresponds
            to a given flow direction. Shape is (N, 3, H, W) or (3, H, W) depending on the input.
438
439
440
441
442
    """

    if flow.dtype != torch.float:
        raise ValueError(f"Flow should be of dtype torch.float, got {flow.dtype}.")

443
444
445
    orig_shape = flow.shape
    if flow.ndim == 3:
        flow = flow[None]  # Add batch dim
446

447
448
449
450
    if flow.ndim != 4 or flow.shape[1] != 2:
        raise ValueError(f"Input flow should have shape (2, H, W) or (N, 2, H, W), got {orig_shape}.")

    max_norm = torch.sum(flow ** 2, dim=1).sqrt().max()
451
452
    epsilon = torch.finfo((flow).dtype).eps
    normalized_flow = flow / (max_norm + epsilon)
453
454
455
456
457
    img = _normalized_flow_to_image(normalized_flow)

    if len(orig_shape) == 3:
        img = img[0]  # Remove batch dim
    return img
458
459
460
461
462
463


@torch.no_grad()
def _normalized_flow_to_image(normalized_flow: torch.Tensor) -> torch.Tensor:

    """
464
    Converts a batch of normalized flow to an RGB image.
465
466

    Args:
467
        normalized_flow (torch.Tensor): Normalized flow tensor of shape (N, 2, H, W)
468
    Returns:
469
       img (Tensor(N, 3, H, W)): Flow visualization image of dtype uint8.
470
471
    """

472
    N, _, H, W = normalized_flow.shape
473
474
475
    device = normalized_flow.device
    flow_image = torch.zeros((N, 3, H, W), dtype=torch.uint8, device=device)
    colorwheel = _make_colorwheel().to(device)  # shape [55x3]
476
    num_cols = colorwheel.shape[0]
477
478
    norm = torch.sum(normalized_flow ** 2, dim=1).sqrt()
    a = torch.atan2(-normalized_flow[:, 1, :, :], -normalized_flow[:, 0, :, :]) / torch.pi
479
480
481
482
483
484
485
486
487
488
489
490
    fk = (a + 1) / 2 * (num_cols - 1)
    k0 = torch.floor(fk).to(torch.long)
    k1 = k0 + 1
    k1[k1 == num_cols] = 0
    f = fk - k0

    for c in range(colorwheel.shape[1]):
        tmp = colorwheel[:, c]
        col0 = tmp[k0] / 255.0
        col1 = tmp[k1] / 255.0
        col = (1 - f) * col0 + f * col1
        col = 1 - norm * (1 - col)
491
        flow_image[:, c, :, :] = torch.floor(255 * col)
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
    return flow_image


def _make_colorwheel() -> torch.Tensor:
    """
    Generates a color wheel for optical flow visualization as presented in:
    Baker et al. "A Database and Evaluation Methodology for Optical Flow" (ICCV, 2007)
    URL: http://vision.middlebury.edu/flow/flowEval-iccv07.pdf.

    Returns:
        colorwheel (Tensor[55, 3]): Colorwheel Tensor.
    """

    RY = 15
    YG = 6
    GC = 4
    CB = 11
    BM = 13
    MR = 6

    ncols = RY + YG + GC + CB + BM + MR
    colorwheel = torch.zeros((ncols, 3))
    col = 0

    # RY
    colorwheel[0:RY, 0] = 255
    colorwheel[0:RY, 1] = torch.floor(255 * torch.arange(0, RY) / RY)
    col = col + RY
    # YG
    colorwheel[col : col + YG, 0] = 255 - torch.floor(255 * torch.arange(0, YG) / YG)
    colorwheel[col : col + YG, 1] = 255
    col = col + YG
    # GC
    colorwheel[col : col + GC, 1] = 255
    colorwheel[col : col + GC, 2] = torch.floor(255 * torch.arange(0, GC) / GC)
    col = col + GC
    # CB
    colorwheel[col : col + CB, 1] = 255 - torch.floor(255 * torch.arange(CB) / CB)
    colorwheel[col : col + CB, 2] = 255
    col = col + CB
    # BM
    colorwheel[col : col + BM, 2] = 255
    colorwheel[col : col + BM, 0] = torch.floor(255 * torch.arange(0, BM) / BM)
    col = col + BM
    # MR
    colorwheel[col : col + MR, 2] = 255 - torch.floor(255 * torch.arange(MR) / MR)
    colorwheel[col : col + MR, 0] = 255
    return colorwheel


542
def _generate_color_palette(num_objects: int):
543
    palette = torch.tensor([2 ** 25 - 1, 2 ** 15 - 1, 2 ** 21 - 1])
544
    return [tuple((i * palette) % 255) for i in range(num_objects)]
545
546


Kai Zhang's avatar
Kai Zhang committed
547
def _log_api_usage_once(obj: Any) -> None:
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564

    """
    Logs API usage(module and name) within an organization.
    In a large ecosystem, it's often useful to track the PyTorch and
    TorchVision APIs usage. This API provides the similar functionality to the
    logging module in the Python stdlib. It can be used for debugging purpose
    to log which methods are used and by default it is inactive, unless the user
    manually subscribes a logger via the `SetAPIUsageLogger method <https://github.com/pytorch/pytorch/blob/eb3b9fe719b21fae13c7a7cf3253f970290a573e/c10/util/Logging.cpp#L114>`_.
    Please note it is triggered only once for the same API call within a process.
    It does not collect any data from open-source users since it is no-op by default.
    For more information, please refer to
    * PyTorch note: https://pytorch.org/docs/stable/notes/large_scale_deployments.html#api-usage-logging;
    * Logging policy: https://github.com/pytorch/vision/issues/5052;

    Args:
        obj (class instance or method): an object to extract info from.
    """
Kai Zhang's avatar
Kai Zhang committed
565
    if not obj.__module__.startswith("torchvision"):
566
        return
Kai Zhang's avatar
Kai Zhang committed
567
568
569
570
    name = obj.__class__.__name__
    if isinstance(obj, FunctionType):
        name = obj.__name__
    torch._C._log_api_usage_once(f"{obj.__module__}.{name}")