Breaking News: Grepper is joining You.com. Read the official announcement!

Accessing an image tensor with floating point indices

Arthur Elskens answered on May 16, 2024 Popularity 1/10 Helpfulness 1/10

answer Accessing an image tensor with floating point indices

Accessing an image tensor with floating point indices

Comment

import torch
import torch.nn.functional as F

# Initialize the tensor
bs, nc, h, w = 1, 3, 256, 256
image_tensor = torch.randn(bs, nc, h, w)

# Define the floating point coordinates as tensors
x = torch.tensor([5.6, 20.3, 150.8])
y = torch.tensor([128.4, 50.7, 200.9])

# Normalize the coordinates to the range [-1, 1]
# The coordinates should be in the range [-1, 1] for grid_sample
x_normalized = (x / (w - 1)) * 2 - 1
y_normalized = (y / (h - 1)) * 2 - 1

# Stack and reshape the coordinates to create a grid tensor
# The shape should be (1, 1, N, 2) where N is the number of points
grid = torch.stack((x_normalized, y_normalized), dim=-1)
grid = grid.unsqueeze(0).unsqueeze(0)

# Sample the points using grid_sample
# grid_sample expects a 4D tensor of shape (N, C, H, W)
interpolated_pixel_values = F.grid_sample(image_tensor, grid, mode='bilinear', align_corners=True)

# The result will be a tensor of shape (1, 3, 1, N)
# We can squeeze it to get rid of the extra dimensions
interpolated_pixel_values = interpolated_pixel_values.squeeze().permute(1, 0)

print(interpolated_pixel_values)

Popularity 1/10 Helpfulness 1/10 Language python

Source: discuss.pytorch.org

Tags: floating-point image indices location python pytorch subpixel tensor

Link to this answer
Share Copy Link

Contributed on May 16 2024

Arthur Elskens

0 Answers Avg Quality 2/10

Accessing an image tensor with floating point indices

Contents

More Related Answers

Accessing an image tensor with floating point indices

Grepper

Documentation

Social

Legal

Contact

Oops, You will need to install Grepper and log-in to perform this action.