PaddleOCR/ppocr/modeling/backbones/rec_vary_vit.py

# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
#
# 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.

import math
from functools import partial
from typing import Optional, Tuple, Type

import numpy as np
import paddle
import paddle.nn as nn
import paddle.nn.functional as F
from paddle.nn.initializer import (
    Constant,
    KaimingUniform,
    Normal,
    TruncatedNormal,
    XavierUniform,
)
from ppocr.modeling.backbones.rec_donut_swin import DonutSwinModelOutput

zeros_ = Constant(value=0.0)
ones_ = Constant(value=1.0)
kaiming_normal_ = KaimingUniform(nonlinearity="relu")
trunc_normal_ = TruncatedNormal(std=0.02)
xavier_uniform_ = XavierUniform()


class MLPBlock(nn.Layer):
    def __init__(
        self,
        embedding_dim: int,
        mlp_dim: int,
        act: Type[nn.Layer] = nn.GELU,
    ) -> None:
        super().__init__()
        self.lin1 = nn.Linear(embedding_dim, mlp_dim)
        self.lin2 = nn.Linear(mlp_dim, embedding_dim)
        self.act = act()

    def forward(self, x):
        return self.lin2(self.act(self.lin1(x)))


# From https://github.com/facebookresearch/detectron2/blob/main/detectron2/layers/batch_norm.py # noqa
# Itself from https://github.com/facebookresearch/ConvNeXt/blob/d1fa8f6fef0a165b27399986cc2bdacc92777e40/models/convnext.py#L119  # noqa
class LayerNorm2d(nn.Layer):
    def __init__(self, num_channels: int, epsilon: float = 1e-6) -> None:
        super().__init__()
        self.weight = paddle.create_parameter([num_channels], dtype="float32")
        ones_(self.weight)
        self.bias = paddle.create_parameter([num_channels], dtype="float32")
        zeros_(self.bias)
        self.epsilon = epsilon

    def forward(self, x):
        u = x.mean(1, keepdim=True)
        s = (x - u).pow(2).mean(1, keepdim=True)
        x = (x - u) / paddle.sqrt(s + self.epsilon)
        x = self.weight[:, None, None] * x + self.bias[:, None, None]
        return x


# This class and its supporting functions below lightly adapted from the ViTDet backbone available at: https://github.com/facebookresearch/detectron2/blob/main/detectron2/modeling/backbone/vit.py # noqa
class ImageEncoderViT(nn.Layer):
    def __init__(
        self,
        img_size: int = 1024,
        patch_size: int = 16,
        in_chans: int = 3,
        embed_dim: int = 768,
        depth: int = 12,
        num_heads: int = 12,
        mlp_ratio: float = 4.0,
        out_chans: int = 256,
        qkv_bias: bool = True,
        norm_layer: Type[nn.Layer] = nn.LayerNorm,
        act_layer: Type[nn.Layer] = nn.GELU,
        use_abs_pos: bool = True,
        use_rel_pos: bool = False,
        rel_pos_zero_init: bool = True,
        window_size: int = 0,
        global_attn_indexes: Tuple[int, ...] = (),
        is_formula: bool = False,
    ) -> None:
        """
        Args:
            img_size (int): Input image size.
            patch_size (int): Patch size.
            in_chans (int): Number of input image channels.
            embed_dim (int): Patch embedding dimension.
            depth (int): Depth of ViT.
            num_heads (int): Number of attention heads in each ViT block.
            mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
            qkv_bias (bool): If True, add a learnable bias to query, key, value.
            norm_layer (nn.Layer): Normalization layer.
            act_layer (nn.Layer): Activation layer.
            use_abs_pos (bool): If True, use absolute positional embeddings.
            use_rel_pos (bool): If True, add relative positional embeddings to the attention map.
            rel_pos_zero_init (bool): If True, zero initialize relative positional parameters.
            window_size (int): Window size for window attention blocks.
            global_attn_indexes (list): Indexes for blocks using global attention.
        """
        super().__init__()
        self.img_size = img_size

        self.patch_embed = PatchEmbed(
            kernel_size=(patch_size, patch_size),
            stride=(patch_size, patch_size),
            in_chans=in_chans,
            embed_dim=embed_dim,
        )

        self.pos_embed = None
        if use_abs_pos:
            # Initialize absolute positional embedding with pretrain image size.
            self.pos_embed = paddle.create_parameter(
                shape=(1, img_size // patch_size, img_size // patch_size, embed_dim),
                dtype="float32",
            )
            zeros_(self.pos_embed)

        self.blocks = nn.LayerList()
        for i in range(depth):
            block = Block(
                dim=embed_dim,
                num_heads=num_heads,
                mlp_ratio=mlp_ratio,
                qkv_bias=qkv_bias,
                norm_layer=norm_layer,
                act_layer=act_layer,
                use_rel_pos=use_rel_pos,
                rel_pos_zero_init=rel_pos_zero_init,
                window_size=window_size if i not in global_attn_indexes else 0,
                input_size=(img_size // patch_size, img_size // patch_size),
            )
            self.blocks.append(block)

        self.neck = nn.Sequential(
            nn.Conv2D(
                embed_dim,
                out_chans,
                kernel_size=1,
                bias_attr=False,
            ),
            LayerNorm2d(out_chans),
            nn.Conv2D(
                out_chans,
                out_chans,
                kernel_size=3,
                padding=1,
                bias_attr=False,
            ),
            LayerNorm2d(out_chans),
        )

        self.net_2 = nn.Conv2D(
            256, 512, kernel_size=3, stride=2, padding=1, bias_attr=False
        )
        self.net_3 = nn.Conv2D(
            512, 1024, kernel_size=3, stride=2, padding=1, bias_attr=False
        )
        self.is_formula = is_formula

    def forward(self, x):
        x = self.patch_embed(x)
        if self.pos_embed is not None:
            x = x + self.pos_embed
        for blk in self.blocks:
            x = blk(x)
        x = self.neck(x.transpose([0, 3, 1, 2]))
        x = self.net_2(x)
        if self.is_formula:
            x = self.net_3(x)
        return x


class Block(nn.Layer):
    """Transformer blocks with support of window attention and residual propagation blocks"""

    def __init__(
        self,
        dim: int,
        num_heads: int,
        mlp_ratio: float = 4.0,
        qkv_bias: bool = True,
        norm_layer: Type[nn.Layer] = nn.LayerNorm,
        act_layer: Type[nn.Layer] = nn.GELU,
        use_rel_pos: bool = False,
        rel_pos_zero_init: bool = True,
        window_size: int = 0,
        input_size: Optional[Tuple[int, int]] = None,
    ) -> None:
        """
        Args:
            dim (int): Number of input channels.
            num_heads (int): Number of attention heads in each ViT block.
            mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
            qkv_bias (bool): If True, add a learnable bias to query, key, value.
            norm_layer (nn.Layer): Normalization layer.
            act_layer (nn.Layer): Activation layer.
            use_rel_pos (bool): If True, add relative positional embeddings to the attention map.
            rel_pos_zero_init (bool): If True, zero initialize relative positional parameters.
            window_size (int): Window size for window attention blocks. If it equals 0, then
                use global attention.
            input_size (tuple(int, int) or None): Input resolution for calculating the relative
                positional parameter size.
        """
        super().__init__()
        self.norm1 = norm_layer(dim)
        self.attn = Attention(
            dim,
            num_heads=num_heads,
            qkv_bias=qkv_bias,
            use_rel_pos=use_rel_pos,
            rel_pos_zero_init=rel_pos_zero_init,
            input_size=input_size if window_size == 0 else (window_size, window_size),
        )

        self.norm2 = norm_layer(dim)
        self.mlp = MLPBlock(
            embedding_dim=dim, mlp_dim=int(dim * mlp_ratio), act=act_layer
        )

        self.window_size = window_size

    def forward(self, x):
        shortcut = x

        x = self.norm1(x)
        # Window partition
        if self.window_size > 0:
            H, W = x.shape[1], x.shape[2]
            x, pad_hw = window_partition(x, self.window_size)
        x = self.attn(x)
        # Reverse window partition
        if self.window_size > 0:
            x = window_unpartition(x, self.window_size, pad_hw, (H, W))
        x = shortcut + x
        x = x + self.mlp(self.norm2(x))

        return x


class Attention(nn.Layer):
    """Multi-head Attention block with relative position embeddings."""

    def __init__(
        self,
        dim: int,
        num_heads: int = 8,
        qkv_bias: bool = True,
        use_rel_pos: bool = False,
        rel_pos_zero_init: bool = True,
        input_size: Optional[Tuple[int, int]] = None,
    ) -> None:
        """
        Args:
            dim (int): Number of input channels.
            num_heads (int): Number of attention heads.
            qkv_bias (bool):  If True, add a learnable bias to query, key, value.
            rel_pos (bool): If True, add relative positional embeddings to the attention map.
            rel_pos_zero_init (bool): If True, zero initialize relative positional parameters.
            input_size (tuple(int, int) or None): Input resolution for calculating the relative
                positional parameter size.
        """
        super().__init__()
        self.num_heads = num_heads
        head_dim = dim // num_heads
        self.scale = head_dim**-0.5

        self.qkv = nn.Linear(dim, dim * 3, bias_attr=qkv_bias)
        self.proj = nn.Linear(dim, dim)

        self.use_rel_pos = use_rel_pos
        if self.use_rel_pos:
            assert (
                input_size is not None
            ), "Input size must be provided if using relative positional encoding."
            # initialize relative positional embeddings
            self.rel_pos_h = paddle.create_parameter(
                [2 * input_size[0] - 1, head_dim], dtype="float32"
            )
            zeros_(self.rel_pos_h)
            self.rel_pos_w = paddle.create_parameter(
                [2 * input_size[1] - 1, head_dim], dtype="float32"
            )
            zeros_(self.rel_pos_w)

    def forward(self, x):

        B, H, W, _ = x.shape
        qkv = (
            self.qkv(x)
            .reshape([B, H * W, 3, self.num_heads, -1])
            .transpose([2, 0, 3, 1, 4])
        )
        q, k, v = qkv.reshape([3, B * self.num_heads, H * W, -1]).unbind(0)
        attn = (q * self.scale) @ k.transpose([0, 2, 1])

        if self.use_rel_pos:
            attn = add_decomposed_rel_pos(
                attn, q, self.rel_pos_h, self.rel_pos_w, (H, W), (H, W)
            )
        attn = F.softmax(attn, axis=-1)
        x = (
            (attn @ v)
            .reshape([B, self.num_heads, H, W, -1])
            .transpose([0, 2, 3, 1, 4])
            .reshape([B, H, W, -1])
        )
        x = self.proj(x)

        return x


def window_partition(x, window_size: int):
    """
    Partition into non-overlapping windows with padding if needed.
    Args:
        x (tensor): input tokens with [B, H, W, C].
        window_size (int): window size.

    Returns:
        windows: windows after partition with [B * num_windows, window_size, window_size, C].
        (Hp, Wp): padded height and width before partition
    """
    B, H, W, C = x.shape

    pad_h = (window_size - H % window_size) % window_size
    pad_w = (window_size - W % window_size) % window_size
    if pad_h > 0 or pad_w > 0:
        x = F.pad(x, (0, 0, 0, pad_w, 0, pad_h, 0, 0))
    Hp, Wp = H + pad_h, W + pad_w

    x = x.reshape(
        [B, Hp // window_size, window_size, Wp // window_size, window_size, C]
    )
    windows = x.transpose([0, 1, 3, 2, 4, 5]).reshape([-1, window_size, window_size, C])
    return windows, (Hp, Wp)


def window_unpartition(
    windows, window_size: int, pad_hw: Tuple[int, int], hw: Tuple[int, int]
):
    """
    Window unpartition into original sequences and removing padding.
    Args:
        windows (tensor): input tokens with [B * num_windows, window_size, window_size, C].
        window_size (int): window size.
        pad_hw (Tuple): padded height and width (Hp, Wp).
        hw (Tuple): original height and width (H, W) before padding.

    Returns:
        x: unpartitioned sequences with [B, H, W, C].
    """
    Hp, Wp = pad_hw
    H, W = hw
    B = windows.shape[0] // (Hp * Wp // window_size // window_size)
    x = windows.reshape(
        [B, Hp // window_size, Wp // window_size, window_size, window_size, -1]
    )
    x = x.transpose([0, 1, 3, 2, 4, 5]).contiguous().reshape([B, Hp, Wp, -1])

    if Hp > H or Wp > W:
        x = x[:, :H, :W, :].contiguous()
    return x


def get_rel_pos(q_size: int, k_size: int, rel_pos):
    """
    Get relative positional embeddings according to the relative positions of
        query and key sizes.
    Args:
        q_size (int): size of query q.
        k_size (int): size of key k.
        rel_pos (Tensor): relative position embeddings (L, C).

    Returns:
        Extracted positional embeddings according to relative positions.
    """
    max_rel_dist = int(2 * max(q_size, k_size) - 1)
    # Interpolate rel pos if needed.
    if rel_pos.shape[0] != max_rel_dist:
        # Interpolate rel pos.
        rel_pos_resized = F.interpolate(
            rel_pos.reshape(1, rel_pos.shape[0], -1).transpose(0, 2, 1),
            size=max_rel_dist,
            mode="linear",
        )
        rel_pos_resized = rel_pos_resized.reshape(-1, max_rel_dist).transpose(1, 0)
    else:
        rel_pos_resized = rel_pos

    # Scale the coords with short length if shapes for q and k are different.
    q_coords = paddle.arange(q_size)[:, None] * max(k_size / q_size, 1.0)
    k_coords = paddle.arange(k_size)[None, :] * max(q_size / k_size, 1.0)
    relative_coords = (q_coords - k_coords) + (k_size - 1) * max(q_size / k_size, 1.0)

    return rel_pos_resized[relative_coords.cast(paddle.int64)]


def add_decomposed_rel_pos(
    attn,
    q,
    rel_pos_h,
    rel_pos_w,
    q_size: Tuple[int, int],
    k_size: Tuple[int, int],
):
    """
    Calculate decomposed Relative Positional Embeddings from :paper:`mvitv2`.
    https://github.com/facebookresearch/mvit/blob/19786631e330df9f3622e5402b4a419a263a2c80/mvit/models/attention.py   # noqa B950
    Args:
        attn (Tensor): attention map.
        q (Tensor): query q in the attention layer with shape (B, q_h * q_w, C).
        rel_pos_h (Tensor): relative position embeddings (Lh, C) for height axis.
        rel_pos_w (Tensor): relative position embeddings (Lw, C) for width axis.
        q_size (Tuple): spatial sequence size of query q with (q_h, q_w).
        k_size (Tuple): spatial sequence size of key k with (k_h, k_w).

    Returns:
        attn (Tensor): attention map with added relative positional embeddings.
    """
    q_h, q_w = q_size
    k_h, k_w = k_size
    Rh = get_rel_pos(q_h, k_h, rel_pos_h)
    Rw = get_rel_pos(q_w, k_w, rel_pos_w)

    B, _, dim = q.shape
    r_q = q.reshape([B, q_h, q_w, dim])
    rel_h = paddle.einsum("bhwc,hkc->bhwk", r_q, Rh)
    rel_w = paddle.einsum("bhwc,wkc->bhwk", r_q, Rw)

    attn = (
        attn.reshape([B, q_h, q_w, k_h, k_w])
        + rel_h[:, :, :, :, None]
        + rel_w[:, :, :, None, :]
    ).reshape([B, q_h * q_w, k_h * k_w])

    return attn


class PatchEmbed(nn.Layer):
    """
    Image to Patch Embedding.
    """

    def __init__(
        self,
        kernel_size: Tuple[int, int] = (16, 16),
        stride: Tuple[int, int] = (16, 16),
        padding: Tuple[int, int] = (0, 0),
        in_chans: int = 3,
        embed_dim: int = 768,
    ) -> None:
        """
        Args:
            kernel_size (Tuple): kernel size of the projection layer.
            stride (Tuple): stride of the projection layer.
            padding (Tuple): padding size of the projection layer.
            in_chans (int): Number of input image channels.
            embed_dim (int): Patch embedding dimension.
        """
        super().__init__()

        self.proj = nn.Conv2D(
            in_chans,
            embed_dim,
            kernel_size=kernel_size,
            stride=stride,
            padding=padding,
            weight_attr=True,
            bias_attr=True,
        )

    def forward(self, x):
        x = self.proj(x)
        # B C H W -> B H W C
        x = x.transpose([0, 2, 3, 1])
        return x


def _build_vary(
    encoder_embed_dim,
    encoder_depth,
    encoder_num_heads,
    encoder_global_attn_indexes,
    image_size,
    is_formula=False,
):
    prompt_embed_dim = 256
    vit_patch_size = 16
    image_embedding_size = image_size // vit_patch_size
    image_encoder = ImageEncoderViT(
        depth=encoder_depth,
        embed_dim=encoder_embed_dim,
        img_size=image_size,
        mlp_ratio=4,
        norm_layer=partial(paddle.nn.LayerNorm, epsilon=1e-6),
        num_heads=encoder_num_heads,
        patch_size=vit_patch_size,
        qkv_bias=True,
        use_rel_pos=True,
        global_attn_indexes=encoder_global_attn_indexes,
        window_size=14,
        out_chans=prompt_embed_dim,
        is_formula=is_formula,
    )
    return image_encoder


class Vary_VIT_B(nn.Layer):
    def __init__(
        self,
        in_channels=3,
        image_size=768,
        encoder_embed_dim=768,
        encoder_depth=12,
        encoder_num_heads=12,
        encoder_global_attn_indexes=[2, 5, 8, 11],
    ):
        super().__init__()

        self.vision_tower_high = _build_vary(
            encoder_embed_dim=768,
            encoder_depth=12,
            encoder_num_heads=12,
            encoder_global_attn_indexes=[2, 5, 8, 11],
            image_size=image_size,
        )

        self.out_channels = 1024

    def forward(self, input_data):
        pixel_values = input_data
        num_channels = pixel_values.shape[1]
        if num_channels == 1:
            pixel_values = paddle.repeat_interleave(pixel_values, repeats=3, axis=1)
        cnn_feature = self.vision_tower_high(pixel_values)
        cnn_feature = cnn_feature.flatten(2).transpose([0, 2, 1])
        return cnn_feature


class Vary_VIT_B_Formula(nn.Layer):
    def __init__(
        self,
        in_channels=3,
        image_size=768,
        encoder_embed_dim=768,
        encoder_depth=12,
        encoder_num_heads=12,
        encoder_global_attn_indexes=[2, 5, 8, 11],
    ):
        """
        Vary_VIT_B_Formula
        Args:
            in_channels (int): Number of input channels. Default is 3 (for RGB images).
            image_size (int): Size of the input image. Default is 768.
            encoder_embed_dim (int): Dimension of the encoder's embedding. Default is 768.
            encoder_depth (int): Number of layers (depth) in the encoder. Default is 12.
            encoder_num_heads (int): Number of attention heads in the encoder. Default is 12.
            encoder_global_attn_indexes (list): List of indices specifying which encoder layers use global attention. Default is [2, 5, 8, 11].
        Returns:
            model: nn.Layer. Specific `Vary_VIT_B_Formula` model with defined architecture.
        """
        super(Vary_VIT_B_Formula, self).__init__()

        self.vision_tower_high = _build_vary(
            encoder_embed_dim=encoder_embed_dim,
            encoder_depth=encoder_depth,
            encoder_num_heads=encoder_num_heads,
            encoder_global_attn_indexes=[2, 5, 8, 11],
            image_size=image_size,
            is_formula=True,
        )
        self.mm_projector_vary = nn.Linear(1024, 1024)
        self.out_channels = 1024

    def forward(self, input_data):
        if self.training:
            pixel_values, label, attention_mask = input_data
        else:
            if isinstance(input_data, list):
                pixel_values = input_data[0]
            else:
                pixel_values = input_data
        num_channels = pixel_values.shape[1]
        if num_channels == 1:
            pixel_values = paddle.repeat_interleave(pixel_values, repeats=3, axis=1)

        cnn_feature = self.vision_tower_high(pixel_values)
        cnn_feature = cnn_feature.flatten(2).transpose([0, 2, 1])

        cnn_feature = self.mm_projector_vary(cnn_feature)
        donut_swin_output = DonutSwinModelOutput(
            last_hidden_state=cnn_feature,
            pooler_output=None,
            hidden_states=None,
            attentions=None,
            reshaped_hidden_states=None,
        )
        if self.training:
            return donut_swin_output, label, attention_mask
        else:
            return donut_swin_output