position_encoding.py

# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import math
import paddle
import paddle.nn as nn

from ppdet.core.workspace import register, serializable


@register
@serializable
class PositionEmbedding(nn.Layer):
    def __init__(self,
                 num_pos_feats=128,
                 temperature=10000,
                 normalize=True,
                 scale=None,
                 embed_type='sine',
                 num_embeddings=50):
        super(PositionEmbedding, self).__init__()
        assert embed_type in ['sine', 'learned']

        self.embed_type = embed_type
        if self.embed_type == 'sine':
            self.num_pos_feats = num_pos_feats
            self.temperature = temperature
            self.normalize = normalize
            if scale is not None and normalize is False:
                raise ValueError("normalize should be True if scale is passed")
            if scale is None:
                scale = 2 * math.pi
            self.scale = scale
        elif self.embed_type == 'learned':
            self.row_embed = nn.Embedding(num_embeddings, num_pos_feats)
            self.col_embed = nn.Embedding(num_embeddings, num_pos_feats)
        else:
            raise ValueError(f"not supported {self.embed_type}")

    def forward(self, mask):
        """
        Args:
            mask (Tensor): [B, H, W]
        Returns:
            pos (Tensor): [B, C, H, W]
        """
        assert mask.dtype == paddle.bool
        if self.embed_type == 'sine':
            mask = mask.astype('float32')
            y_embed = mask.cumsum(1, dtype='float32')
            x_embed = mask.cumsum(2, dtype='float32')
            if self.normalize:
                eps = 1e-6
                y_embed = y_embed / (y_embed[:, -1:, :] + eps) * self.scale
                x_embed = x_embed / (x_embed[:, :, -1:] + eps) * self.scale

            dim_t = 2 * (paddle.arange(self.num_pos_feats) //
                         2).astype('float32')
            dim_t = self.temperature**(dim_t / self.num_pos_feats)

            pos_x = x_embed.unsqueeze(-1) / dim_t
            pos_y = y_embed.unsqueeze(-1) / dim_t
            pos_x = paddle.stack(
                (pos_x[:, :, :, 0::2].sin(), pos_x[:, :, :, 1::2].cos()),
                axis=4).flatten(3)
            pos_y = paddle.stack(
                (pos_y[:, :, :, 0::2].sin(), pos_y[:, :, :, 1::2].cos()),
                axis=4).flatten(3)
            pos = paddle.concat((pos_y, pos_x), axis=3).transpose([0, 3, 1, 2])
            return pos
        elif self.embed_type == 'learned':
            h, w = mask.shape[-2:]
            i = paddle.arange(w)
            j = paddle.arange(h)
            x_emb = self.col_embed(i)
            y_emb = self.row_embed(j)
            pos = paddle.concat(
                [
                    x_emb.unsqueeze(0).repeat(h, 1, 1),
                    y_emb.unsqueeze(1).repeat(1, w, 1),
                ],
                axis=-1).transpose([2, 0, 1]).unsqueeze(0).tile(mask.shape[0],
                                                                1, 1, 1)
            return pos
        else:
            raise ValueError(f"not supported {self.embed_type}")