roi_align.py

import functools
from typing import List, Union

import torch
import torch._dynamo
import torch.fx
from torch import nn, Tensor
from torch._dynamo.utils import is_compile_supported
from torch.jit.annotations import BroadcastingList2
from torch.nn.modules.utils import _pair
from torchvision.extension import _assert_has_ops, _has_ops

from ..utils import _log_api_usage_once
from ._utils import check_roi_boxes_shape, convert_boxes_to_roi_format


def lazy_compile(**compile_kwargs):
    """Lazily wrap a function with torch.compile on the first call

    This avoids eagerly importing dynamo.
    """

    def decorate_fn(fn):
        @functools.wraps(fn)
        def compile_hook(*args, **kwargs):
            compiled_fn = torch.compile(fn, **compile_kwargs)
            globals()[fn.__name__] = functools.wraps(fn)(compiled_fn)
            return compiled_fn(*args, **kwargs)

        return compile_hook

    return decorate_fn


# NB: all inputs are tensors
def _bilinear_interpolate(
    input,  # [N, C, H, W]
    roi_batch_ind,  # [K]
    y,  # [K, PH, IY]
    x,  # [K, PW, IX]
    ymask,  # [K, IY]
    xmask,  # [K, IX]
):
    _, channels, height, width = input.size()

    # deal with inverse element out of feature map boundary
    y = y.clamp(min=0)
    x = x.clamp(min=0)
    y_low = y.int()
    x_low = x.int()
    y_high = torch.where(y_low >= height - 1, height - 1, y_low + 1)
    y_low = torch.where(y_low >= height - 1, height - 1, y_low)
    y = torch.where(y_low >= height - 1, y.to(input.dtype), y)

    x_high = torch.where(x_low >= width - 1, width - 1, x_low + 1)
    x_low = torch.where(x_low >= width - 1, width - 1, x_low)
    x = torch.where(x_low >= width - 1, x.to(input.dtype), x)

    ly = y - y_low
    lx = x - x_low
    hy = 1.0 - ly
    hx = 1.0 - lx

    # do bilinear interpolation, but respect the masking!
    # TODO: It's possible the masking here is unnecessary if y and
    # x were clamped appropriately; hard to tell
    def masked_index(
        y,  # [K, PH, IY]
        x,  # [K, PW, IX]
    ):
        if ymask is not None:
            assert xmask is not None
            y = torch.where(ymask[:, None, :], y, 0)
            x = torch.where(xmask[:, None, :], x, 0)
        return input[
            roi_batch_ind[:, None, None, None, None, None],
            torch.arange(channels, device=input.device)[None, :, None, None, None, None],
            y[:, None, :, None, :, None],  # prev [K, PH, IY]
            x[:, None, None, :, None, :],  # prev [K, PW, IX]
        ]  # [K, C, PH, PW, IY, IX]

    v1 = masked_index(y_low, x_low)
    v2 = masked_index(y_low, x_high)
    v3 = masked_index(y_high, x_low)
    v4 = masked_index(y_high, x_high)

    # all ws preemptively [K, C, PH, PW, IY, IX]
    def outer_prod(y, x):
        return y[:, None, :, None, :, None] * x[:, None, None, :, None, :]

    w1 = outer_prod(hy, hx)
    w2 = outer_prod(hy, lx)
    w3 = outer_prod(ly, hx)
    w4 = outer_prod(ly, lx)

    val = w1 * v1 + w2 * v2 + w3 * v3 + w4 * v4
    return val


# TODO: this doesn't actually cache
# TODO: main library should make this easier to do
def maybe_cast(tensor):
    if torch.is_autocast_enabled() and tensor.is_cuda and tensor.dtype != torch.double:
        return tensor.float()
    else:
        return tensor


# This is a pure Python and differentiable implementation of roi_align.  When
# run in eager mode, it uses a lot of memory, but when compiled it has
# acceptable memory usage.  The main point of this implementation is that
# its backwards is deterministic.
# It is transcribed directly off of the roi_align CUDA kernel, see
# https://dev-discuss.pytorch.org/t/a-pure-python-implementation-of-roi-align-that-looks-just-like-its-cuda-kernel/1266
@lazy_compile(dynamic=True)
def _roi_align(input, rois, spatial_scale, pooled_height, pooled_width, sampling_ratio, aligned):
    orig_dtype = input.dtype

    input = maybe_cast(input)
    rois = maybe_cast(rois)

    _, _, height, width = input.size()

    ph = torch.arange(pooled_height, device=input.device)  # [PH]
    pw = torch.arange(pooled_width, device=input.device)  # [PW]

    # input: [N, C, H, W]
    # rois: [K, 5]

    roi_batch_ind = rois[:, 0].int()  # [K]
    offset = 0.5 if aligned else 0.0
    roi_start_w = rois[:, 1] * spatial_scale - offset  # [K]
    roi_start_h = rois[:, 2] * spatial_scale - offset  # [K]
    roi_end_w = rois[:, 3] * spatial_scale - offset  # [K]
    roi_end_h = rois[:, 4] * spatial_scale - offset  # [K]

    roi_width = roi_end_w - roi_start_w  # [K]
    roi_height = roi_end_h - roi_start_h  # [K]
    if not aligned:
        roi_width = torch.clamp(roi_width, min=1.0)  # [K]
        roi_height = torch.clamp(roi_height, min=1.0)  # [K]

    bin_size_h = roi_height / pooled_height  # [K]
    bin_size_w = roi_width / pooled_width  # [K]

    exact_sampling = sampling_ratio > 0

    roi_bin_grid_h = sampling_ratio if exact_sampling else torch.ceil(roi_height / pooled_height)  # scalar or [K]
    roi_bin_grid_w = sampling_ratio if exact_sampling else torch.ceil(roi_width / pooled_width)  # scalar or [K]

    """
    iy, ix = dims(2)
    """

    if exact_sampling:
        count = max(roi_bin_grid_h * roi_bin_grid_w, 1)  # scalar
        iy = torch.arange(roi_bin_grid_h, device=input.device)  # [IY]
        ix = torch.arange(roi_bin_grid_w, device=input.device)  # [IX]
        ymask = None
        xmask = None
    else:
        count = torch.clamp(roi_bin_grid_h * roi_bin_grid_w, min=1)  # [K]
        # When doing adaptive sampling, the number of samples we need to do
        # is data-dependent based on how big the ROIs are.  This is a bit
        # awkward because first-class dims can't actually handle this.
        # So instead, we inefficiently suppose that we needed to sample ALL
        # the points and mask out things that turned out to be unnecessary
        iy = torch.arange(height, device=input.device)  # [IY]
        ix = torch.arange(width, device=input.device)  # [IX]
        ymask = iy[None, :] < roi_bin_grid_h[:, None]  # [K, IY]
        xmask = ix[None, :] < roi_bin_grid_w[:, None]  # [K, IX]

    def from_K(t):
        return t[:, None, None]

    y = (
        from_K(roi_start_h)
        + ph[None, :, None] * from_K(bin_size_h)
        + (iy[None, None, :] + 0.5).to(input.dtype) * from_K(bin_size_h / roi_bin_grid_h)
    )  # [K, PH, IY]
    x = (
        from_K(roi_start_w)
        + pw[None, :, None] * from_K(bin_size_w)
        + (ix[None, None, :] + 0.5).to(input.dtype) * from_K(bin_size_w / roi_bin_grid_w)
    )  # [K, PW, IX]
    val = _bilinear_interpolate(input, roi_batch_ind, y, x, ymask, xmask)  # [K, C, PH, PW, IY, IX]

    # Mask out samples that weren't actually adaptively needed
    if not exact_sampling:
        val = torch.where(ymask[:, None, None, None, :, None], val, 0)
        val = torch.where(xmask[:, None, None, None, None, :], val, 0)

    output = val.sum((-1, -2))  # remove IY, IX ~> [K, C, PH, PW]
    if isinstance(count, torch.Tensor):
        output /= count[:, None, None, None]
    else:
        output /= count

    output = output.to(orig_dtype)

    return output


@torch.fx.wrap
def roi_align(
    input: Tensor,
    boxes: Union[Tensor, List[Tensor]],
    output_size: BroadcastingList2[int],
    spatial_scale: float = 1.0,
    sampling_ratio: int = -1,
    aligned: bool = False,
) -> Tensor:
    """
    Performs Region of Interest (RoI) Align operator with average pooling, as described in Mask R-CNN.

    Args:
        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 box coordinates to
            the input coordinates. For example, if your boxes are defined on the scale
            of a 224x224 image and your input is a 112x112 feature map (resulting from a 0.5x scaling of
            the original image), you'll want to set this to 0.5. 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:
        Tensor[K, C, output_size[0], output_size[1]]: The pooled RoIs.
    """
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(roi_align)
    check_roi_boxes_shape(boxes)
    rois = boxes
    output_size = _pair(output_size)
    if not isinstance(rois, torch.Tensor):
        rois = convert_boxes_to_roi_format(rois)
    if not torch.jit.is_scripting():
        if (
            not _has_ops() or (torch.are_deterministic_algorithms_enabled() and (input.is_cuda or input.is_mps))
        ) and is_compile_supported(input.device.type):
            return _roi_align(input, rois, spatial_scale, output_size[0], output_size[1], sampling_ratio, aligned)
    _assert_has_ops()
    return torch.ops.torchvision.roi_align(
        input, rois, spatial_scale, output_size[0], output_size[1], sampling_ratio, aligned
    )


class RoIAlign(nn.Module):
    """
    See :func:`roi_align`.
    """

    def __init__(
        self,
        output_size: BroadcastingList2[int],
        spatial_scale: float,
        sampling_ratio: int,
        aligned: bool = False,
    ):
        super().__init__()
        _log_api_usage_once(self)
        self.output_size = output_size
        self.spatial_scale = spatial_scale
        self.sampling_ratio = sampling_ratio
        self.aligned = aligned

    def forward(self, input: Tensor, rois: Union[Tensor, List[Tensor]]) -> Tensor:
        return roi_align(input, rois, self.output_size, self.spatial_scale, self.sampling_ratio, self.aligned)

    def __repr__(self) -> str:
        s = (
            f"{self.__class__.__name__}("
            f"output_size={self.output_size}"
            f", spatial_scale={self.spatial_scale}"
            f", sampling_ratio={self.sampling_ratio}"
            f", aligned={self.aligned}"
            f")"
        )
        return s