# torchvision.ops¶

torchvision.ops implements operators that are specific for Computer Vision.

Note

All operators have native support for TorchScript.

torchvision.ops.nms(boxes: torch.Tensor, scores: torch.Tensor, iou_threshold: float) → torch.Tensor[source]

Performs non-maximum suppression (NMS) on the boxes according to their intersection-over-union (IoU).

NMS iteratively removes lower scoring boxes which have an IoU greater than iou_threshold with another (higher scoring) box.

If multiple boxes have the exact same score and satisfy the IoU criterion with respect to a reference box, the selected box is not guaranteed to be the same between CPU and GPU. This is similar to the behavior of argsort in PyTorch when repeated values are present.

Parameters
• boxes (Tensor[N, 4])) – boxes to perform NMS on. They are expected to be in (x1, y1, x2, y2) format with 0 <= x1 < x2 and 0 <= y1 < y2.

• scores (Tensor[N]) – scores for each one of the boxes

• iou_threshold (float) – discards all overlapping boxes with IoU > iou_threshold

Returns

int64 tensor with the indices of the elements that have been kept by NMS, sorted in decreasing order of scores

Return type

Tensor

torchvision.ops.batched_nms(boxes: torch.Tensor, scores: torch.Tensor, idxs: torch.Tensor, iou_threshold: float) → torch.Tensor[source]

Performs non-maximum suppression in a batched fashion.

Each index value correspond to a category, and NMS will not be applied between elements of different categories.

Parameters
• boxes (Tensor[N, 4]) – boxes where NMS will be performed. They are expected to be in (x1, y1, x2, y2) format with 0 <= x1 < x2 and 0 <= y1 < y2.

• scores (Tensor[N]) – scores for each one of the boxes

• idxs (Tensor[N]) – indices of the categories for each one of the boxes.

• iou_threshold (float) – discards all overlapping boxes with IoU > iou_threshold

Returns

int64 tensor with the indices of the elements that have been kept by NMS, sorted in decreasing order of scores

Return type

Tensor

torchvision.ops.remove_small_boxes(boxes: torch.Tensor, min_size: float) → torch.Tensor[source]

Remove boxes which contains at least one side smaller than min_size.

Parameters
• boxes (Tensor[N, 4]) – boxes in (x1, y1, x2, y2) format with 0 <= x1 < x2 and 0 <= y1 < y2.

• min_size (float) – minimum size

Returns

indices of the boxes that have both sides larger than min_size

Return type

Tensor[K]

torchvision.ops.clip_boxes_to_image(boxes: torch.Tensor, size: Tuple[int, int]) → torch.Tensor[source]

Clip boxes so that they lie inside an image of size size.

Parameters
• boxes (Tensor[N, 4]) – boxes in (x1, y1, x2, y2) format with 0 <= x1 < x2 and 0 <= y1 < y2.

• size (Tuple[height, width]) – size of the image

Returns

clipped boxes

Return type

Tensor[N, 4]

torchvision.ops.box_convert(boxes: torch.Tensor, in_fmt: str, out_fmt: str) → torch.Tensor[source]

Converts boxes from given in_fmt to out_fmt. Supported in_fmt and out_fmt are:

‘xyxy’: boxes are represented via corners, x1, y1 being top left and x2, y2 being bottom right. This is the format that torchvision utilities expect.

‘xywh’ : boxes are represented via corner, width and height, x1, y2 being top left, w, h being width and height.

‘cxcywh’ : boxes are represented via centre, width and height, cx, cy being center of box, w, h being width and height.

Parameters
• boxes (Tensor[N, 4]) – boxes which will be converted.

• in_fmt (str) – Input format of given boxes. Supported formats are [‘xyxy’, ‘xywh’, ‘cxcywh’].

• out_fmt (str) – Output format of given boxes. Supported formats are [‘xyxy’, ‘xywh’, ‘cxcywh’]

Returns

Boxes into converted format.

Return type

Tensor[N, 4]

torchvision.ops.box_area(boxes: torch.Tensor) → torch.Tensor[source]

Computes the area of a set of bounding boxes, which are specified by their (x1, y1, x2, y2) coordinates.

Parameters

boxes (Tensor[N, 4]) – boxes for which the area will be computed. They are expected to be in (x1, y1, x2, y2) format with 0 <= x1 < x2 and 0 <= y1 < y2.

Returns

the area for each box

Return type

Tensor[N]

torchvision.ops.box_iou(boxes1: torch.Tensor, boxes2: torch.Tensor) → torch.Tensor[source]

Return intersection-over-union (Jaccard index) between two sets of boxes.

Both sets of boxes are expected to be in (x1, y1, x2, y2) format with 0 <= x1 < x2 and 0 <= y1 < y2.

Parameters
• boxes1 (Tensor[N, 4]) – first set of boxes

• boxes2 (Tensor[M, 4]) – second set of boxes

Returns

the NxM matrix containing the pairwise IoU values for every element in boxes1 and boxes2

Return type

Tensor[N, M]

torchvision.ops.generalized_box_iou(boxes1: torch.Tensor, boxes2: torch.Tensor) → torch.Tensor[source]

Return generalized intersection-over-union (Jaccard index) between two sets of boxes.

Both sets of boxes are expected to be in (x1, y1, x2, y2) format with 0 <= x1 < x2 and 0 <= y1 < y2.

Parameters
• boxes1 (Tensor[N, 4]) – first set of boxes

• boxes2 (Tensor[M, 4]) – second set of boxes

Returns

the NxM matrix containing the pairwise generalized IoU values for every element in boxes1 and boxes2

Return type

Tensor[N, M]

torchvision.ops.roi_align(input: torch.Tensor, boxes: torch.Tensor, output_size: None, spatial_scale: float = 1.0, sampling_ratio: int = -1, aligned: bool = False) → torch.Tensor[source]

Performs Region of Interest (RoI) Align operator with average pooling, as described in Mask R-CNN.

Parameters
• input (Tensor[N, C, H, W]) – The input tensor, i.e. a batch with N elements. Each element contains C feature maps of dimensions H x W. If the tensor is quantized, we expect a batch size of N == 1.

• boxes (Tensor[K, 5] or List[Tensor[L, 4]]) – the box coordinates in (x1, y1, x2, y2) format where the regions will be taken from. The coordinate must satisfy 0 <= x1 < x2 and 0 <= y1 < y2. If a single Tensor is passed, then the first column should contain the index of the corresponding element in the batch, i.e. a number in [0, N - 1]. If a list of Tensors is passed, then each Tensor will correspond to the boxes for an element i in the batch.

• output_size (int or Tuple[int, int]) – the size of the output (in bins or pixels) after the pooling is performed, as (height, width).

• spatial_scale (float) – a scaling factor that maps the input coordinates to the box coordinates. Default: 1.0

• sampling_ratio (int) – number of sampling points in the interpolation grid used to compute the output value of each pooled output bin. If > 0, then exactly sampling_ratio x sampling_ratio sampling points per bin are used. If <= 0, then an adaptive number of grid points are used (computed as ceil(roi_width / output_width), and likewise for height). Default: -1

• aligned (bool) – If False, use the legacy implementation. If True, pixel shift the box coordinates it by -0.5 for a better alignment with the two neighboring pixel indices. This version is used in Detectron2

Returns

The pooled RoIs.

Return type

Tensor[K, C, output_size[0], output_size[1]]

torchvision.ops.ps_roi_align(input: torch.Tensor, boxes: torch.Tensor, output_size: int, spatial_scale: float = 1.0, sampling_ratio: int = -1) → torch.Tensor[source]

Performs Position-Sensitive Region of Interest (RoI) Align operator mentioned in Light-Head R-CNN.

Parameters
• input (Tensor[N, C, H, W]) – The input tensor, i.e. a batch with N elements. Each element contains C feature maps of dimensions H x W.

• boxes (Tensor[K, 5] or List[Tensor[L, 4]]) – the box coordinates in (x1, y1, x2, y2) format where the regions will be taken from. The coordinate must satisfy 0 <= x1 < x2 and 0 <= y1 < y2. If a single Tensor is passed, then the first column should contain the index of the corresponding element in the batch, i.e. a number in [0, N - 1]. If a list of Tensors is passed, then each Tensor will correspond to the boxes for an element i in the batch.

• output_size (int or Tuple[int, int]) – the size of the output (in bins or pixels) after the pooling is performed, as (height, width).

• spatial_scale (float) – a scaling factor that maps the input coordinates to the box coordinates. Default: 1.0

• sampling_ratio (int) – number of sampling points in the interpolation grid used to compute the output value of each pooled output bin. If > 0, then exactly sampling_ratio x sampling_ratio sampling points per bin are used. If <= 0, then an adaptive number of grid points are used (computed as ceil(roi_width / output_width), and likewise for height). Default: -1

Returns

The pooled RoIs

Return type

Tensor[K, C / (output_size[0] * output_size[1]), output_size[0], output_size[1]]

torchvision.ops.roi_pool(input: torch.Tensor, boxes: torch.Tensor, output_size: None, spatial_scale: float = 1.0) → torch.Tensor[source]

Performs Region of Interest (RoI) Pool operator described in Fast R-CNN

Parameters
• input (Tensor[N, C, H, W]) – The input tensor, i.e. a batch with N elements. Each element contains C feature maps of dimensions H x W.

• boxes (Tensor[K, 5] or List[Tensor[L, 4]]) – the box coordinates in (x1, y1, x2, y2) format where the regions will be taken from. The coordinate must satisfy 0 <= x1 < x2 and 0 <= y1 < y2. If a single Tensor is passed, then the first column should contain the index of the corresponding element in the batch, i.e. a number in [0, N - 1]. If a list of Tensors is passed, then each Tensor will correspond to the boxes for an element i in the batch.

• output_size (int or Tuple[int, int]) – the size of the output after the cropping is performed, as (height, width)

• spatial_scale (float) – a scaling factor that maps the input coordinates to the box coordinates. Default: 1.0

Returns

The pooled RoIs.

Return type

Tensor[K, C, output_size[0], output_size[1]]

torchvision.ops.ps_roi_pool(input: torch.Tensor, boxes: torch.Tensor, output_size: int, spatial_scale: float = 1.0) → torch.Tensor[source]

Performs Position-Sensitive Region of Interest (RoI) Pool operator described in R-FCN

Parameters
• input (Tensor[N, C, H, W]) – The input tensor, i.e. a batch with N elements. Each element contains C feature maps of dimensions H x W.

• boxes (Tensor[K, 5] or List[Tensor[L, 4]]) – the box coordinates in (x1, y1, x2, y2) format where the regions will be taken from. The coordinate must satisfy 0 <= x1 < x2 and 0 <= y1 < y2. If a single Tensor is passed, then the first column should contain the index of the corresponding element in the batch, i.e. a number in [0, N - 1]. If a list of Tensors is passed, then each Tensor will correspond to the boxes for an element i in the batch.

• output_size (int or Tuple[int, int]) – the size of the output (in bins or pixels) after the pooling is performed, as (height, width).

• spatial_scale (float) – a scaling factor that maps the input coordinates to the box coordinates. Default: 1.0

Returns

The pooled RoIs.

Return type

Tensor[K, C / (output_size[0] * output_size[1]), output_size[0], output_size[1]]

torchvision.ops.deform_conv2d(input: torch.Tensor, offset: torch.Tensor, weight: torch.Tensor, bias: Optional[torch.Tensor] = None, stride: Tuple[int, int] = (1, 1), padding: Tuple[int, int] = (0, 0), dilation: Tuple[int, int] = (1, 1), mask: Optional[torch.Tensor] = None) → torch.Tensor[source]

Performs Deformable Convolution v2, described in Deformable ConvNets v2: More Deformable, Better Results if mask is not None and Performs Deformable Convolution, described in Deformable Convolutional Networks if mask is None.

Parameters
• input (Tensor[batch_size, in_channels, in_height, in_width]) – input tensor

• offset (Tensor[batch_size, 2 * offset_groups * kernel_height * kernel_width, out_height, out_width]) – offsets to be applied for each position in the convolution kernel.

• weight (Tensor[out_channels, in_channels // groups, kernel_height, kernel_width]) – convolution weights, split into groups of size (in_channels // groups)

• bias (Tensor[out_channels]) – optional bias of shape (out_channels,). Default: None

• stride (int or Tuple[int, int]) – distance between convolution centers. Default: 1

• padding (int or Tuple[int, int]) – height/width of padding of zeroes around each image. Default: 0

• dilation (int or Tuple[int, int]) – the spacing between kernel elements. Default: 1

• mask (Tensor[batch_size, offset_groups * kernel_height * kernel_width, out_height, out_width]) – masks to be applied for each position in the convolution kernel. Default: None

Returns

result of convolution

Return type

Tensor[batch_sz, out_channels, out_h, out_w]

Examples::
>>> input = torch.rand(4, 3, 10, 10)
>>> kh, kw = 3, 3
>>> weight = torch.rand(5, 3, kh, kw)
>>> # offset and mask should have the same spatial size as the output
>>> # of the convolution. In this case, for an input of 10, stride of 1
>>> # and kernel size of 3, without padding, the output size is 8
>>> offset = torch.rand(4, 2 * kh * kw, 8, 8)
>>> mask = torch.rand(4, kh * kw, 8, 8)
>>> out = deform_conv2d(input, offset, weight, mask=mask)
>>> print(out.shape)
>>> # returns
>>>  torch.Size([4, 5, 8, 8])

torchvision.ops.sigmoid_focal_loss(inputs: torch.Tensor, targets: torch.Tensor, alpha: float = 0.25, gamma: float = 2, reduction: str = 'none')[source]

Original implementation from https://github.com/facebookresearch/fvcore/blob/master/fvcore/nn/focal_loss.py . Loss used in RetinaNet for dense detection: https://arxiv.org/abs/1708.02002.

Parameters
• inputs – A float tensor of arbitrary shape. The predictions for each example.

• targets – A float tensor with the same shape as inputs. Stores the binary classification label for each element in inputs (0 for the negative class and 1 for the positive class).

• alpha – (optional) Weighting factor in range (0,1) to balance positive vs negative examples or -1 for ignore. Default = 0.25

• gamma – Exponent of the modulating factor (1 - p_t) to balance easy vs hard examples.

• reduction – ‘none’ | ‘mean’ | ‘sum’ ‘none’: No reduction will be applied to the output. ‘mean’: The output will be averaged. ‘sum’: The output will be summed.

Returns

Loss tensor with the reduction option applied.

class torchvision.ops.RoIAlign(output_size: None, spatial_scale: float, sampling_ratio: int, aligned: bool = False)[source]
class torchvision.ops.PSRoIAlign(output_size: int, spatial_scale: float, sampling_ratio: int)[source]
class torchvision.ops.RoIPool(output_size: None, spatial_scale: float)[source]
class torchvision.ops.PSRoIPool(output_size: int, spatial_scale: float)[source]
class torchvision.ops.DeformConv2d(in_channels: int, out_channels: int, kernel_size: int, stride: int = 1, padding: int = 0, dilation: int = 1, groups: int = 1, bias: bool = True)[source]
class torchvision.ops.MultiScaleRoIAlign(featmap_names: List[str], output_size: Union[int, Tuple[int], List[int]], sampling_ratio: int, *, canonical_scale: int = 224, canonical_level: int = 4)[source]

Multi-scale RoIAlign pooling, which is useful for detection with or without FPN.

It infers the scale of the pooling via the heuristics specified in eq. 1 of the Feature Pyramid Network paper. They keyword-only parameters canonical_scale and canonical_level correspond respectively to 224 and k0=4 in eq. 1, and have the following meaning: canonical_level is the target level of the pyramid from which to pool a region of interest with w x h = canonical_scale x canonical_scale.

Parameters
• featmap_names (List[str]) – the names of the feature maps that will be used for the pooling.

• output_size (List[Tuple[int, int]] or List[int]) – output size for the pooled region

• sampling_ratio (int) – sampling ratio for ROIAlign

• canonical_scale (int, optional) – canonical_scale for LevelMapper

• canonical_level (int, optional) – canonical_level for LevelMapper

Examples:

>>> m = torchvision.ops.MultiScaleRoIAlign(['feat1', 'feat3'], 3, 2)
>>> i = OrderedDict()
>>> i['feat1'] = torch.rand(1, 5, 64, 64)
>>> i['feat2'] = torch.rand(1, 5, 32, 32)  # this feature won't be used in the pooling
>>> i['feat3'] = torch.rand(1, 5, 16, 16)
>>> # create some random bounding boxes
>>> boxes = torch.rand(6, 4) * 256; boxes[:, 2:] += boxes[:, :2]
>>> # original image size, before computing the feature maps
>>> image_sizes = [(512, 512)]
>>> output = m(i, [boxes], image_sizes)
>>> print(output.shape)
>>> torch.Size([6, 5, 3, 3])

class torchvision.ops.FeaturePyramidNetwork(in_channels_list: List[int], out_channels: int, extra_blocks: Optional[torchvision.ops.feature_pyramid_network.ExtraFPNBlock] = None)[source]

Module that adds a FPN from on top of a set of feature maps. This is based on “Feature Pyramid Network for Object Detection”.

The feature maps are currently supposed to be in increasing depth order.

The input to the model is expected to be an OrderedDict[Tensor], containing the feature maps on top of which the FPN will be added.

Parameters
• in_channels_list (list[int]) – number of channels for each feature map that is passed to the module

• out_channels (int) – number of channels of the FPN representation

• extra_blocks (ExtraFPNBlock or None) – if provided, extra operations will be performed. It is expected to take the fpn features, the original features and the names of the original features as input, and returns a new list of feature maps and their corresponding names

Examples:

>>> m = torchvision.ops.FeaturePyramidNetwork([10, 20, 30], 5)
>>> # get some dummy data
>>> x = OrderedDict()
>>> x['feat0'] = torch.rand(1, 10, 64, 64)
>>> x['feat2'] = torch.rand(1, 20, 16, 16)
>>> x['feat3'] = torch.rand(1, 30, 8, 8)
>>> # compute the FPN on top of x
>>> output = m(x)
>>> print([(k, v.shape) for k, v in output.items()])
>>> # returns
>>>   [('feat0', torch.Size([1, 5, 64, 64])),
>>>    ('feat2', torch.Size([1, 5, 16, 16])),
>>>    ('feat3', torch.Size([1, 5, 8, 8]))]