adaptive_avgpool2d

fn adaptive_avgpool2d(
    x: &Tensor<T>,
    output_size: [i64; 2]
) -> Result<Tensor<C>, TensorError>

Performs an adaptive avg pooling operation on the input tensor, automatically determining the kernel size and stride to produce the specified output dimensions.

Parameters:

x: Input tensor with shape [batch_size, height, width, channels]

output_size: Desired output spatial dimensions as [out_height, out_width]

Returns:

Tensor with type C

Examples:

use hpt::{
    common::TensorInfo,
    error::TensorError,
    ops::{FloatOutPooling, Random},
    Tensor,
};

fn main() -> Result<(), TensorError> {
    // [batch_size, height, width, channels]
    let input = Tensor::<f32>::randn([1, 32, 32, 16])?;

    // Perform adaptive avg pooling to get 16x16 output
    let output = input.adaptive_avgpool2d([16, 16])?;

    println!("Output shape: {:?}", output.shape()); // [1, 16, 16, 16]

    // Resize to a different output size
    let output2 = input.adaptive_avgpool2d([8, 8])?;

    println!("Output2 shape: {:?}", output2.shape()); // [1, 8, 8, 16]
    Ok(())
}

Backend Support

Backend	Supported
CPU	✅
Cuda	❌