Document feedback

Question document fragment

When a question document fragment contains a formula, it is displayed as a space.

Submission type

issue

It's a little complicated...

I'd like to ask someone.

PR

Just a small problem.

I can fix it online!

Please select the submission type

Problem type

Specifications and Common Mistakes

- Specifications and Common Mistakes:

- Misspellings or punctuation mistakes,incorrect formulas, abnormal display.

- Incorrect links, empty cells, or wrong formats.

- Chinese characters in English context.

- Minor inconsistencies between the UI and descriptions.

- Low writing fluency that does not affect understanding.

- Incorrect version numbers, including software package names and version numbers on the UI.

Usability

- Usability:

- Incorrect or missing key steps.

- Missing main function descriptions, keyword explanation, necessary prerequisites, or precautions.

- Ambiguous descriptions, unclear reference, or contradictory context.

- Unclear logic, such as missing classifications, items, and steps.

Correctness

- Correctness:

- Technical principles, function descriptions, supported platforms, parameter types, or exceptions inconsistent with that of software implementation.

- Incorrect schematic or architecture diagrams.

- Incorrect commands or command parameters.

- Incorrect code.

- Commands inconsistent with the functions.

- Wrong screenshots.

- Sample code running error, or running results inconsistent with the expectation.

Risk Warnings

- Risk Warnings:

- Lack of risk warnings for operations that may damage the system or important data.

Content Compliance

- Content Compliance:

- Contents that may violate applicable laws and regulations or geo-cultural context-sensitive words and expressions.

- Copyright infringement.

Please select the type of question

Problem description

Describe the bug so that we can quickly locate the problem.

Document feedback

Function Differences with torchvision.ops.deform_conv2d

torchvision.ops.deform_conv2d

class torchvision.ops.deform_conv2d(
    input,
    offset,
    weight,
    bias=None,
    stride=(1, 1),
    padding=(0, 0),
    dilations=(1, 1),
    mask=None
)

For more information, see torchvision.ops.deform_conv2d.

mindspore.ops.deformable_conv2d

class mindspore.ops.deformable_conv2d(
    x,
    weight,
    offsets,
    kernel_size,
    strides,
    padding,
    bias=None,
    dilations=(1, 1, 1, 1),
    groups=1,
    deformable_groups=1,
    modulated=True
)

For more information, see mindspore.ops.deformable_conv2d.

Differences

PyTorch: Parameters offsets is a 4D tensor of x-y coordinates offset. With the format “NCHW”, the shape is $(b a t c h, d e f o r m a b l e \underset{―}{} g r o u p s \times H_{f} \times W_{f} \times 2, H_{out}, W_{out})$ . Note the C dimension is stored in the order of $(d e f o r m a b l e \underset{―}{} g r o u p s, H_{f}, W_{f}, (o f f s e t \underset{―}{} y, o f f s e t \underset{―}{} x))$ . Parameters mask is a 4D tensor of mask. With the format “NCHW”, the shape is $(b a t c h, d e f o r m a b l e \underset{―}{} g r o u p s \times H_{f} \times W_{f} \times 1, H_{out}, W_{out})$ . Note the C dimension is stored in the order of $(d e f o r m a b l e \underset{―}{} g r o u p s, H_{f}, W_{f}, m a s k)$ .

MindSpore: Parameters offsets is a 4D tensor of x-y coordinates offset and mask. With the format “NCHW”, the shape is $(b a t c h, 3 \times d e f o r m a b l e \underset{―}{} g r o u p s \times H_{f} \times W_{f}, H_{out}, W_{out})$ . Note the C dimension is stored in the order of $((o f f s e t \underset{―}{} x, o f f s e t \underset{―}{} y, m a s k), d e f o r m a b l e \underset{―}{} g r o u p s, H_{f}, W_{f})$ .

Categories	Subcategories	PyTorch	MindSpore	Difference
Parameters	Parameter 1	input	x	Same function, different parameter names
	Parameter 2	offset	offsets	MindSpore parameters offsets is a 4D tensor of x-y coordinates offset and mask
	Parameter 3	weight	weight	-
	Parameter 4	-	kernel_size	Pytorch does not have this parameter
	Parameter 5	mask	-	MindSpore does not have this parameter
	Parameter 6	bias	bias	-
	Parameter 7	stride	strides	Same function, different parameter names
	Parameter 8	padding	padding	-
	Parameter 9	dilations	dilations	-
	Parameter 10	-	groups	Pytorch does not have this parameter
	Parameter 11	-	deformable_groups	Pytorch does not have this parameter
	Parameter 12	-	modulated	Pytorch does not have this parameter

Code Example# PyTorch
import torch
from torch import tensor
import numpy as np
from torchvision.ops import deform_conv2d
np.random.seed(1)
kh, kw = 1, 1
batch = 1
deformable_groups = 1
stride_h, stride_w = 1, 1
dilation_h, dilation_w = 1, 1
pad_h, pad_w = 0, 0
x_h, x_w = 1, 2
out_h = (x_h + 2 * pad_h - dilation_h * (kh - 1) - 1) // stride_h + 1
out_w = (x_w + 2 * pad_w - dilation_w * (kw - 1) - 1) // stride_w + 1

x = np.random.randn(batch, 64, x_h, x_w).astype(np.float32)
weight = np.random.randn(batch, 64, kh, kw).astype(np.float32)
offsets_x = np.random.randn(batch, 1, deformable_groups, kh, kw, out_h, out_w).astype(np.float32)
offsets_y = np.random.randn(batch, 1, deformable_groups, kh, kw, out_h, out_w).astype(np.float32)
mask = np.random.randn(batch, 1, deformable_groups, kh, kw, out_h, out_w).astype(np.float32)

offsets = np.concatenate((offsets_y, offsets_x), axis=1)
offsets = offsets.transpose(0, 2, 3, 4, 1, 5, 6)
offsets = offsets.reshape((batch, 2 * deformable_groups * kh * kw, out_h, out_w))
mask = mask.transpose(0, 2, 3, 4, 1, 5, 6)
mask = mask.reshape((batch, 1 * deformable_groups * kh * kw, out_h, out_w))
x = torch.from_numpy(x.copy().astype(np.float32))
weight = torch.from_numpy(weight.copy().astype(np.float32))
offsets = torch.from_numpy(offsets.copy().astype(np.float32))
mask = torch.from_numpy(mask.copy().astype(np.float32))
output = deform_conv2d(x, offsets, weight, stride=(stride_h, stride_w), padding=(pad_h, pad_w), dilation=(dilation_h, dilation_w), mask=mask)
print(output)
# tensor([[[[-0.0022,  0.0000]]]])

# MindSpore
import mindspore
from mindspore import Tensor
import mindspore.nn as nn
import numpy as np
from mindspore.ops import deformable_conv2d
import mindspore.ops as ops
np.random.seed(1)
kh, kw = 1, 1
batch = 1
deformable_groups = 1
stride_h, stride_w = 1, 1
dilation_h, dilation_w = 1, 1
pad_h, pad_w = 0, 0
x_h, x_w = 1, 2
out_h = (x_h + 2 * pad_h - dilation_h * (kh - 1) - 1) // stride_h + 1
out_w = (x_w + 2 * pad_w - dilation_w * (kw - 1) - 1) // stride_w + 1

x = np.random.randn(batch, 64, x_h, x_w).astype(np.float32)
weight = np.random.randn(batch, 64, kh, kw).astype(np.float32)
offsets_x = np.random.randn(batch, 1, deformable_groups, kh, kw, out_h, out_w).astype(np.float32)
offsets_y = np.random.randn(batch, 1, deformable_groups, kh, kw, out_h, out_w).astype(np.float32)
mask = np.random.randn(batch, 1, deformable_groups, kh, kw, out_h, out_w).astype(np.float32)

offsets = np.concatenate((offsets_x, offsets_y, mask), axis=1)
offsets = offsets.reshape((batch, 3 * deformable_groups * kh * kw, out_h, out_w))

x = Tensor(x)
weight = Tensor(weight)
offsets = Tensor(offsets)
output = ops.deformable_conv2d(x, weight, offsets, (kh, kw), (1, 1, stride_h, stride_w,), (pad_h, pad_h, pad_w, pad_w), dilations=(1, 1, dilation_h, dilation_w))
print(output)
# [[[[-0.00220442  0.        ]]]]