nexaai 1.0.29__cp310-cp310-macosx_14_0_universal2.whl

nexaai/mlx_backend/vlm/modeling/models/multi_modality/sam.py

@@ -0,0 +1,557 @@
+import copy
+from dataclasses import dataclass
+from typing import List, Optional, Tuple, Type, Union
+
+import mlx.core as mx
+import mlx.nn as nn
+import numpy as np
+from PIL import Image
+from PIL.Image import Resampling
+
+
+@dataclass
+class SAMViTCfg:
+    image_size: Union[Tuple[int, int], int] = 1024
+    width: int = 768
+    layers: int = 12
+    heads: int = 12
+    patch_size: int = 16
+    window_size: int = 14
+    prompt_embed_dim: int = 256
+    global_attn_indexes: Union[List[int], Tuple[int]] = (2, 5, 8, 11)
+    downsample_channels: Union[List[int], Tuple[int]] = (512, 1024)
+
+
+class MLPBlock(nn.Module):
+    def __init__(
+        self,
+        embedding_dim: int,
+        mlp_dim: int,
+        act: Type[nn.Module] = 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 __call__(self, x: mx.array):
+        return self.lin2(self.act(self.lin1(x)))
+
+
+def resize_image(image_np, new_size=(96, 96), order=1):
+    """
+    Resize an image with multiple channels using PIL.
+
+    Args:
+    image_np (numpy.ndarray): The input image array of shape (batch, channels, height, width).
+    new_size (tuple): The target size as (height, width).
+    order (int): The order of interpolation (used to determine resampling method).
+
+    Returns:
+    numpy.ndarray: The resized image array in the same format as input.
+    """
+    # Remove batch dimension
+    image_np = np.array(image_np[0])
+
+    # Get dimensions
+    channels, height, width = image_np.shape
+
+    # Choose interpolation method based on order parameter
+    resample_method = Resampling.BILINEAR  # Default to bilinear
+    if order == 0:
+        resample_method = Resampling.NEAREST
+    elif order == 2 or order == 3:
+        resample_method = Resampling.BICUBIC
+
+    # Handle different channel configurations
+    if channels == 1:
+        # For single-channel images (grayscale)
+        # Reshape to 2D array (height, width)
+        image_2d = image_np.reshape(height, width)
+
+        # Create PIL image - ensure proper mode and data type conversion
+        pil_image = Image.fromarray(image_2d.astype(np.float32))
+
+        # Resize using PIL (note: PIL takes width, height order)
+        resized_pil = pil_image.resize(
+            (new_size[1], new_size[0]), resample=resample_method
+        )
+
+        # Convert back to numpy array, reshape to add channel dimension
+        resized_np = np.array(resized_pil).reshape((1, new_size[0], new_size[1]))
+    else:
+        # For multi-channel images, process each channel individually
+        resized_channels = []
+
+        for c in range(channels):
+            channel_data = image_np[c]
+            pil_channel = Image.fromarray(channel_data.astype(np.float32))
+            resized_channel = pil_channel.resize(
+                (new_size[1], new_size[0]), resample=resample_method
+            )
+            resized_channels.append(np.array(resized_channel))
+
+        # Stack channels back together
+        resized_np = np.stack(resized_channels, axis=0)
+
+    # Add batch dimension back and convert to mx.array
+    return mx.array(resized_np)[None, :]
+
+
+class SAMEncoder(nn.Module):
+    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.Module] = nn.LayerNorm,
+        act_layer: Type[nn.Module] = nn.GELU,
+        use_abs_pos: bool = True,
+        use_rel_pos: bool = True,
+        rel_pos_zero_init: bool = True,
+        window_size: int = 14,
+        global_attn_indexes: Tuple[int, ...] = (2, 5, 8, 11),
+        downsample_channels: Tuple[int, ...] = (512, 1024),
+    ) -> 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.Module): Normalization layer.
+            act_layer (nn.Module): 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.
+            downsample_channels (list): Channels for downsampling layers.
+        """
+        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,
+        )
+
+        if use_abs_pos:
+            # Initialize absolute positional embedding with pretrain image size.
+            self.pos_embed = mx.zeros(
+                (img_size // patch_size, img_size // patch_size, embed_dim)
+            )[None, :]
+
+        self.blocks = []
+        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.Conv2d(
+                embed_dim,
+                out_chans,
+                kernel_size=1,
+                bias=False,
+            ),
+            nn.LayerNorm(out_chans),
+            nn.Conv2d(
+                out_chans,
+                out_chans,
+                kernel_size=3,
+                padding=1,
+                bias=False,
+            ),
+            nn.LayerNorm(out_chans),
+        ]
+
+        in_channels = out_chans
+        self.downsamples = []
+        for i in range(len(downsample_channels)):
+            out_channels = downsample_channels[i]
+            self.downsamples.append(
+                nn.Conv2d(
+                    in_channels,
+                    out_channels,
+                    kernel_size=3,
+                    stride=2,
+                    padding=1,
+                    bias=False,
+                )
+            )
+            in_channels = out_channels
+
+        self.sam_hd = True
+        if self.sam_hd:
+            self.hd_alpha_downsamples = mx.zeros((1))
+            self.neck_hd = copy.deepcopy(self.neck)
+
+    def __call__(self, x: mx.array):
+        x = self.patch_embed(x)
+        if self.pos_embed is not None:
+            x += self.pos_embed
+
+        global_features = []
+        for _, blk in enumerate(self.blocks):
+            x = blk(x)
+            if self.sam_hd and blk.window_size == 0:
+                global_features.append(x)
+
+        for _, n in enumerate(self.neck):
+            x = n(x)
+
+        x = x.transpose(0, 3, 1, 2)
+        x = resize_image(x)
+
+        x = x.transpose(0, 2, 3, 1)
+
+        for _, downsample in enumerate(self.downsamples):
+            x = downsample(x)
+
+        if self.sam_hd:
+            first_global_feature = global_features[0]
+            for _, n_hd in enumerate(self.neck_hd):
+                first_global_feature = n_hd(first_global_feature)
+
+            first_global_feature = first_global_feature.transpose(0, 3, 1, 2)
+
+            first_global_feature = resize_image(first_global_feature)
+
+            first_global_feature = first_global_feature.transpose(0, 2, 3, 1)
+            for _, downsample in enumerate(self.downsamples):
+                first_global_feature = downsample(first_global_feature)
+
+            x = x + first_global_feature * self.hd_alpha_downsamples
+
+        return x
+
+
+class Block(nn.Module):
+    """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.Module] = nn.LayerNorm,
+        act_layer: Type[nn.Module] = 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.Module): Normalization layer.
+            act_layer (nn.Module): 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 __call__(self, x: mx.array):
+        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.Module):
+    """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=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 = mx.zeros((2 * input_size[0] - 1, head_dim))
+            self.rel_pos_w = mx.zeros((2 * input_size[1] - 1, head_dim))
+
+    def __call__(self, x: mx.array):
+        B, H, W, _ = x.shape
+        x = mx.array(x)
+        # qkv with shape (3, B, nHead, H * W, C)
+        qkv = (
+            self.qkv(x)
+            .reshape(B, H * W, 3, self.num_heads, -1)
+            .transpose(2, 0, 3, 1, 4)
+        )
+        # q, k, v with shape (B * nHead, H * W, C)
+        q, k, v = qkv.reshape(3, B * self.num_heads, H * W, -1)
+
+        def do_attention(q, k, v):
+            attn = (q * self.scale) @ k.transpose(0, -1, -2)
+            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 = mx.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)
+            )
+
+            return x
+
+        x = do_attention(q, k, v)
+        x = self.proj(x)
+
+        return x
+
+
+def window_partition(
+    x: np.ndarray, window_size: int
+) -> Tuple[np.ndarray, Tuple[int, int]]:
+    """
+    Partition into non-overlapping windows with padding if needed.
+    Args:
+        x (ndarray): 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 = np.pad(x, ((0, 0), (0, pad_h), (0, pad_w), (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: np.ndarray,
+    window_size: int,
+    pad_hw: Tuple[int, int],
+    hw: Tuple[int, int],
+):
+    """
+    Window unpartition into original sequences and removing padding.
+    Args:
+        windows (ndarray): 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).reshape(B, Hp, Wp, -1)
+
+    if Hp > H or Wp > W:
+        x = x[:, :H, :W, :]
+    return x
+
+
+def get_rel_pos(q_size: int, k_size: int, rel_pos: np.ndarray) -> np.ndarray:
+    """
+    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 (ndarray): relative position embeddings (L, C).
+
+    Returns:
+        Extracted positional embeddings according to relative positions.
+    """
+    rel_pos = np.array(rel_pos)
+    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 = np.expand_dims(rel_pos, axis=0)
+        rel_pos_resized = np.transpose(rel_pos_resized, (0, 2, 1))
+        rel_pos_resized = np.interp(
+            np.linspace(0, max_rel_dist - 1, num=max_rel_dist),
+            np.arange(rel_pos.shape[0]),
+            rel_pos_resized[0],
+        )
+        rel_pos_resized = np.transpose(rel_pos_resized, (1, 0))
+    else:
+        rel_pos_resized = rel_pos
+
+    # Scale the coords with short length if shapes for q and k are different.
+    q_coords = np.arange(q_size)[:, np.newaxis] * max(k_size / q_size, 1.0)
+    k_coords = np.arange(k_size)[np.newaxis, :] * max(q_size / k_size, 1.0)
+    relative_coords = (q_coords - k_coords) + (k_size - 1) * max(q_size / k_size, 1.0)
+    relative_coords = relative_coords.astype(np.int64)
+    return rel_pos_resized[relative_coords]
+
+
+def add_decomposed_rel_pos(
+    attn: np.ndarray,
+    q: np.ndarray,
+    rel_pos_h: np.ndarray,
+    rel_pos_w: np.ndarray,
+    q_size: Tuple[int, int],
+    k_size: Tuple[int, int],
+) -> np.ndarray:
+    """
+    Calculate decomposed Relative Positional Embeddings from :paper:`mvitv2`.
+    https://github.com/facebookresearch/mvit/blob/19786631e330df9f3622e5402b4a419a263a2c80/mvit/models/attention.py   # noqa B950
+    Args:
+        attn (ndarray): attention map.
+        q (ndarray): query q in the attention layer with shape (B, q_h * q_w, C).
+        rel_pos_h (ndarray): relative position embeddings (Lh, C) for height axis.
+        rel_pos_w (ndarray): 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 (ndarray): 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 = np.einsum("bhwc,hkc->bhwk", r_q, Rh)
+    rel_w = np.einsum("bhwc,wkc->bhwk", r_q, Rw)
+
+    attn = (
+        attn.reshape(B, q_h, q_w, k_h, k_w)
+        + rel_h[:, :, :, :, np.newaxis]
+        + rel_w[:, :, :, np.newaxis, :]
+    ).reshape(B, q_h * q_w, k_h * k_w)
+
+    return attn
+
+
+class PatchEmbed(nn.Module):
+    """
+    Image to Patch Embedding.
+    """
+
+    def __init__(
+        self,
+        kernel_size: Tuple[int, int] = (16, 16),
+        stride: Tuple[int, int] = (16, 16),
+        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
+        )
+
+    def __call__(self, x: mx.array):
+        x = self.proj(x)
+        return x