DenseAV/denseav/featurizers/AudioMAE.py

import math
import os
import warnings
from functools import partial

import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import torchaudio
from timm.models.layers import to_2tuple
from torch.utils.data import Dataset
from torchaudio.functional import resample
import pickle


def _no_grad_trunc_normal_(tensor, mean, std, a, b):
    # Cut & paste from PyTorch official master until it's in a few official releases - RW
    # Method based on https://people.sc.fsu.edu/~jburkardt/presentations/truncated_normal.pdf
    def norm_cdf(x):
        # Computes standard normal cumulative distribution function
        return (1. + math.erf(x / math.sqrt(2.))) / 2.

    if (mean < a - 2 * std) or (mean > b + 2 * std):
        warnings.warn("mean is more than 2 std from [a, b] in nn.init.trunc_normal_. "
                      "The distribution of values may be incorrect.",
                      stacklevel=2)

    with torch.no_grad():
        # Values are generated by using a truncated uniform distribution and
        # then using the inverse CDF for the normal distribution.
        # Get upper and lower cdf values
        l = norm_cdf((a - mean) / std)
        u = norm_cdf((b - mean) / std)

        # Uniformly fill tensor with values from [l, u], then translate to
        # [2l-1, 2u-1].
        tensor.uniform_(2 * l - 1, 2 * u - 1)

        # Use inverse cdf transform for normal distribution to get truncated
        # standard normal
        tensor.erfinv_()

        # Transform to proper mean, std
        tensor.mul_(std * math.sqrt(2.))
        tensor.add_(mean)

        # Clamp to ensure it's in the proper range
        tensor.clamp_(min=a, max=b)
        return tensor


def trunc_normal_(tensor, mean=0., std=1., a=-2., b=2.):
    # type: (Tensor, float, float, float, float) -> Tensor
    r"""Fills the input Tensor with values drawn from a truncated
    normal distribution. The values are effectively drawn from the
    normal distribution :math:`\mathcal{N}(\text{mean}, \text{std}^2)`
    with values outside :math:`[a, b]` redrawn until they are within
    the bounds. The method used for generating the random values works
    best when :math:`a \leq \text{mean} \leq b`.
    Args:
        tensor: an n-dimensional `torch.Tensor`
        mean: the mean of the normal distribution
        std: the standard deviation of the normal distribution
        a: the minimum cutoff value
        b: the maximum cutoff value
    Examples:
        >>> w = torch.empty(3, 5)
        >>> nn.init.trunc_normal_(w)
    """
    return _no_grad_trunc_normal_(tensor, mean, std, a, b)


class Mlp(nn.Module):
    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.):
        super().__init__()
        out_features = out_features or in_features
        hidden_features = hidden_features or in_features
        self.fc1 = nn.Linear(in_features, hidden_features)
        self.act = act_layer()
        self.fc2 = nn.Linear(hidden_features, out_features)
        self.drop = nn.Dropout(drop)

    def forward(self, x):
        x = self.fc1(x)
        x = self.act(x)
        x = self.drop(x)
        x = self.fc2(x)
        x = self.drop(x)
        return x


class Attention(nn.Module):
    def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0.):
        super().__init__()
        self.num_heads = num_heads
        head_dim = dim // num_heads
        # NOTE scale factor was wrong in my original version, can set manually to be compat with prev weights
        self.scale = qk_scale or head_dim ** -0.5

        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
        self.attn_drop = nn.Dropout(attn_drop)
        self.proj = nn.Linear(dim, dim)
        self.proj_drop = nn.Dropout(proj_drop)

    def forward(self, x):
        B, N, C = x.shape
        qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
        q, k, v = qkv[0], qkv[1], qkv[2]  # make torchscript happy (cannot use tensor as tuple)

        attn = (q @ k.transpose(-2, -1)) * self.scale
        attn = attn.softmax(dim=-1)
        attn = self.attn_drop(attn)

        x = (attn @ v).transpose(1, 2).reshape(B, N, C)
        x = self.proj(x)
        x = self.proj_drop(x)
        return x


def drop_path(x, drop_prob: float = 0., training: bool = False):
    """Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).

    This is the same as the DropConnect impl I created for EfficientNet, etc networks, however,
    the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
    See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ... I've opted for
    changing the layer and argument names to 'drop path' rather than mix DropConnect as a layer name and use
    'survival rate' as the argument.

    """
    if drop_prob == 0. or not training:
        return x
    keep_prob = 1 - drop_prob
    shape = (x.shape[0],) + (1,) * (x.ndim - 1)  # work with diff dim tensors, not just 2D ConvNets
    random_tensor = keep_prob + torch.rand(shape, dtype=x.dtype, device=x.device)
    random_tensor.floor_()  # binarize
    output = x.div(keep_prob) * random_tensor
    return output


class DropPath(nn.Module):
    """Drop paths (Stochastic Depth) per sample  (when applied in main path of residual blocks).
    """

    def __init__(self, drop_prob=None):
        super(DropPath, self).__init__()
        self.drop_prob = drop_prob

    def forward(self, x):
        return drop_path(x, self.drop_prob, self.training)


class Block(nn.Module):

    def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop=0., attn_drop=0.,
                 drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm):
        super().__init__()
        self.norm1 = norm_layer(dim)
        self.attn = Attention(
            dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
        # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
        self.norm2 = norm_layer(dim)
        mlp_hidden_dim = int(dim * mlp_ratio)
        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)

    def forward(self, x):
        x = x + self.drop_path(self.attn(self.norm1(x)))
        x = x + self.drop_path(self.mlp(self.norm2(x)))
        return x


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

    def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768):
        super().__init__()
        img_size = to_2tuple(img_size)
        patch_size = to_2tuple(patch_size)
        num_patches = (img_size[1] // patch_size[1]) * (img_size[0] // patch_size[0])
        self.patch_hw = (img_size[1] // patch_size[1], img_size[0] // patch_size[0])
        self.img_size = img_size
        self.patch_size = patch_size
        self.num_patches = num_patches

        self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size)

    def forward(self, x):
        B, C, H, W = x.shape
        # FIXME look at relaxing size constraints
        # assert H == self.img_size[0] and W == self.img_size[1], \
        #    f"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]})."
        x = self.proj(x).flatten(2).transpose(1, 2)
        return x


class HybridEmbed(nn.Module):
    """ CNN Feature Map Embedding
    Extract feature map from CNN, flatten, project to embedding dim.
    """

    def __init__(self, backbone, img_size=224, feature_size=None, in_chans=3, embed_dim=768):
        super().__init__()
        assert isinstance(backbone, nn.Module)
        img_size = to_2tuple(img_size)
        self.img_size = img_size
        self.backbone = backbone
        if feature_size is None:
            with torch.no_grad():
                # FIXME this is hacky, but most reliable way of determining the exact dim of the output feature
                # map for all networks, the feature metadata has reliable channel and stride info, but using
                # stride to calc feature dim requires info about padding of each stage that isn't captured.
                training = backbone.training
                if training:
                    backbone.eval()
                o = self.backbone(torch.zeros(1, in_chans, img_size[0], img_size[1]))[-1]
                feature_size = o.shape[-2:]
                feature_dim = o.shape[1]
                backbone.train(training)
        else:
            feature_size = to_2tuple(feature_size)
            feature_dim = self.backbone.feature_info.channels()[-1]
        self.num_patches = feature_size[0] * feature_size[1]
        self.proj = nn.Linear(feature_dim, embed_dim)

    def forward(self, x):
        x = self.backbone(x)[-1]
        x = x.flatten(2).transpose(1, 2)
        x = self.proj(x)
        return x


class TimmVisionTransformer(nn.Module):
    """ Vision Transformer with support for patch or hybrid CNN input stage
    """

    def __init__(self, img_size=224, patch_size=16, in_chans=3, num_classes=1000, embed_dim=768, depth=12,
                 num_heads=12, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop_rate=0., attn_drop_rate=0.,
                 drop_path_rate=0., hybrid_backbone=None, norm_layer=nn.LayerNorm):
        super().__init__()
        self.num_classes = num_classes
        self.num_features = self.embed_dim = embed_dim  # num_features for consistency with other models

        if hybrid_backbone is not None:
            self.patch_embed = HybridEmbed(
                hybrid_backbone, img_size=img_size, in_chans=in_chans, embed_dim=embed_dim)
        else:
            self.patch_embed = PatchEmbed(
                img_size=img_size, patch_size=patch_size, in_chans=in_chans, embed_dim=embed_dim)
        num_patches = self.patch_embed.num_patches

        self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim))
        self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, embed_dim))
        self.pos_drop = nn.Dropout(p=drop_rate)

        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, depth)]  # stochastic depth decay rule
        self.blocks = nn.ModuleList([
            Block(
                dim=embed_dim, num_heads=num_heads, mlp_ratio=mlp_ratio, qkv_bias=qkv_bias, qk_scale=qk_scale,
                drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[i], norm_layer=norm_layer)
            for i in range(depth)])
        self.norm = norm_layer(embed_dim)

        # NOTE as per official impl, we could have a pre-logits representation dense layer + tanh here
        # self.repr = nn.Linear(embed_dim, representation_size)
        # self.repr_act = nn.Tanh()

        # Classifier head
        self.head = nn.Linear(embed_dim, num_classes) if num_classes > 0 else nn.Identity()

        trunc_normal_(self.pos_embed, std=.02)
        trunc_normal_(self.cls_token, std=.02)
        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            trunc_normal_(m.weight, std=.02)
            if isinstance(m, nn.Linear) and m.bias is not None:
                nn.init.constant_(m.bias, 0)
        elif isinstance(m, nn.LayerNorm):
            nn.init.constant_(m.bias, 0)
            nn.init.constant_(m.weight, 1.0)

    @torch.jit.ignore
    def no_weight_decay(self):
        return {'pos_embed', 'cls_token'}

    def get_classifier(self):
        return self.head

    def reset_classifier(self, num_classes, global_pool=''):
        self.num_classes = num_classes
        self.head = nn.Linear(self.embed_dim, num_classes) if num_classes > 0 else nn.Identity()

    def forward_features(self, x):
        B = x.shape[0]
        x = self.patch_embed(x)

        cls_tokens = self.cls_token.expand(B, -1, -1)  # stole cls_tokens impl from Phil Wang, thanks
        x = torch.cat((cls_tokens, x), dim=1)
        x = x + self.pos_embed
        x = self.pos_drop(x)

        for blk in self.blocks:
            x = blk(x)

        x = self.norm(x)
        return x[:, 0]

    def forward(self, x):
        x = self.forward_features(x)
        x = self.head(x)
        return x


class VisionTransformer(TimmVisionTransformer):
    """ Vision Transformer with support for global average pooling
    """

    def __init__(self, **kwargs):
        super(VisionTransformer, self).__init__(**kwargs)
        norm_layer = kwargs['norm_layer']
        embed_dim = kwargs['embed_dim']
        self.fc_norm = norm_layer(embed_dim)
        del self.norm  # remove the original norm

    def interpolate_pos_encoding(self, x, embed):
        new_patches = x.shape[1]
        old_patches = embed.shape[1]

        w = 8
        h = int(new_patches / w)
        if new_patches == old_patches:
            return embed

        dim = x.shape[-1]
        pos_embed = nn.functional.interpolate(
            embed.reshape(1, 64, 8, dim).permute(0, 3, 1, 2),
            size=(h, w),
            mode='bicubic',
        )
        pos_embed = pos_embed.permute(0, 2, 3, 1).view(1, -1, dim)
        return pos_embed

    def forward(self, x):
        B = x.shape[0]
        x = self.patch_embed(x)

        x = x + self.interpolate_pos_encoding(x, self.pos_embed[:, 1:, :])

        cls_token = self.cls_token + self.pos_embed[:, :1, :]
        cls_tokens = cls_token.expand(B, -1, -1)  # stole cls_tokens impl from Phil Wang, thanks
        x = torch.cat((cls_tokens, x), dim=1)
        x = self.pos_drop(x)

        for blk in self.blocks:
            x = blk(x)

        # x = x[:, 1:, :].mean(dim=1)  # global pool without cls token
        # outcome = self.fc_norm(x)

        return x[:, 1:, :].reshape(B, -1, 8, 768).permute(0, 3, 2, 1), x[:, 0]


class NewPatchEmbed(nn.Module):
    """ Flexible Image to Patch Embedding
    """

    def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768, stride=10):
        super().__init__()
        img_size = to_2tuple(img_size)
        patch_size = to_2tuple(patch_size)
        stride = to_2tuple(stride)
        self.img_size = img_size
        self.patch_size = patch_size
        self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=stride)  # with overlapped patches
        _, _, h, w = self.get_output_shape(img_size)  # n, emb_dim, h, w
        self.patch_hw = (h, w)
        self.num_patches = h * w

    def get_output_shape(self, img_size):
        # todo: don't be lazy..
        return self.proj(torch.randn(1, 1, img_size[0], img_size[1])).shape

    def forward(self, x):
        x = self.proj(x)
        x = x.flatten(2).transpose(1, 2)
        return x


def pca(image_feats_list, dim=3, fit_pca=None):
    from sklearn.decomposition import PCA

    device = image_feats_list[0].device

    def flatten(tensor, target_size=None):
        if target_size is not None and fit_pca is None:
            F.interpolate(tensor, (target_size, target_size), mode="bilinear")
        B, C, H, W = tensor.shape
        return feats.permute(1, 0, 2, 3).reshape(C, B * H * W).permute(1, 0).detach().cpu()

    if len(image_feats_list) > 1 and fit_pca is None:
        target_size = image_feats_list[0].shape[2]
    else:
        target_size = None

    flattened_feats = []
    for feats in image_feats_list:
        flattened_feats.append(flatten(feats, target_size))
    x = torch.cat(flattened_feats, dim=0)

    if fit_pca is None:
        fit_pca = PCA(n_components=dim, svd_solver="arpack").fit(np.nan_to_num(x.detach().numpy()))

    reduced_feats = []
    for feats in image_feats_list:
        x_red = torch.from_numpy(fit_pca.transform(flatten(feats)))
        x_red -= x_red.min(dim=0, keepdim=True).values
        x_red /= x_red.max(dim=0, keepdim=True).values
        B, C, H, W = feats.shape
        reduced_feats.append(x_red.reshape(B, H, W, dim).permute(0, 3, 1, 2).to(device))

    return reduced_feats, fit_pca


class AudiosetDataset(Dataset):
    def __init__(self, audio_conf):
        self.audio_conf = audio_conf
        self.melbins = self.audio_conf.get('num_mel_bins')
        self.dataset = self.audio_conf.get('dataset')
        self.norm_mean = self.audio_conf.get('mean')
        self.norm_std = self.audio_conf.get('std')

        print('Dataset: {}, mean {:.3f} and std {:.3f}'.format(self.dataset, self.norm_mean, self.norm_std))
        print(f'size of dataset {self.__len__()}')

    def _wav2fbank(self, filename):
        sample_rate = 16000
        target_length = 10
        waveform, obs_sr = torchaudio.load(filename)
        waveform = waveform[0]
        if obs_sr != sample_rate:
            waveform = resample(waveform, obs_sr, sample_rate)

        original_length = waveform.shape[0]
        padding = target_length * sample_rate - original_length

        if padding > 0:
            m = torch.nn.ZeroPad2d((0, padding))
            waveform = m(waveform)
        else:
            waveform = waveform[:target_length * sample_rate]


        waveform = waveform - waveform.mean()

        # 498 128, 998, 128
        fbank = torchaudio.compliance.kaldi.fbank(
            waveform.unsqueeze(0),
            htk_compat=True,
            sample_frequency=sample_rate,
            use_energy=False,
            window_type='hanning',
            num_mel_bins=128,
            dither=0.0,
            frame_shift=10)

        normed_fbank = (fbank - self.norm_mean) / (self.norm_std * 2)

        return normed_fbank

    def __getitem__(self, index):
        datum = {"wav": "../../samples/example.wav"}
        fbank = self._wav2fbank(datum['wav'])
        fbank = fbank.transpose(0, 1).unsqueeze(0)  # 1, 128, 1024 (...,freq,time)
        fbank = torch.transpose(fbank.squeeze(), 0, 1)  # time, freq
        # the output fbank shape is [time_frame_num, frequency_bins], e.g., [1024, 128]
        return fbank.unsqueeze(0)

    def __len__(self):
        return 1


class AudioMAE(nn.Module):

    def __init__(self, output_path, finetuned):
        super().__init__()
        # build model
        model = VisionTransformer(
            patch_size=16,
            embed_dim=768,
            depth=12,
            num_heads=12,
            mlp_ratio=4,
            qkv_bias=True,
            norm_layer=partial(nn.LayerNorm, eps=1e-6),
            num_classes=527,
            drop_path_rate=0.1)

        img_size = (1024, 128)  # 1024, 128
        emb_dim = 768
        model.patch_embed = NewPatchEmbed(
            img_size=img_size, patch_size=(16, 16), in_chans=1, embed_dim=emb_dim, stride=16)
        num_patches = model.patch_embed.num_patches
        model.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, emb_dim), requires_grad=False)

        if finetuned:
            fn = "audiomae_finetuned.pth"
        else:
            fn = "audiomae.pth"

        checkpoint = torch.load(os.path.join(output_path, 'models', fn), map_location='cpu')

        checkpoint_model = checkpoint['model']
        state_dict = model.state_dict()
        for k in ['head.weight', 'head.bias']:
            if k in checkpoint_model and checkpoint_model[k].shape != state_dict[k].shape:
                print(f"Removing key {k} from pretrained checkpoint")
                del checkpoint_model[k]
        msg = model.load_state_dict(checkpoint_model, strict=False)
        print(msg)

        model = model.eval()
        self.model = model
        self.config = dict(output_path=output_path, finetuned=finetuned)

    def forward(self, audio, include_cls):
        patch_tokens, cls_token = self.model(audio)

        if include_cls:
            return patch_tokens, cls_token
        else:
            return patch_tokens


if __name__ == '__main__':
    import os

    device = torch.device("cuda:2")

    torch.manual_seed(0)
    np.random.seed(0)

    model = AudioMAE("../../", True).to(device)

    audio_conf_val = {
        'num_mel_bins': 128,
        'target_length': 1024,
        'dataset': "audioset",
        'mode': 'val',
        'mean': -4.2677393,
        'std': 4.5689974,
    }

    dataset = AudiosetDataset(audio_conf=audio_conf_val)

    batch = dataset[0].unsqueeze(0).to(device)

    embeddings = model(batch, include_cls=False)

    import matplotlib.pyplot as plt

    with torch.no_grad():
        [pca_feats], _ = pca([embeddings])
        plt.imshow(pca_feats.cpu().squeeze(0).permute(1, 2, 0))
        plt.show()
        print("here")

    print("here")