PVT v2：超越 Swin 的新型金字塔 ViT

import paddleimport paddle.nn as nnimport paddle.nn.functional as Ffrom functools import partialimport math

trunc_normal_ = nn.initializer.TruncatedNormal(std=.02)
zeros_ = nn.initializer.Constant(value=0.)
ones_ = nn.initializer.Constant(value=1.)
kaiming_normal_ = nn.initializer.KaimingNormal()

def to_2tuple(x):
    return tuple([x] * 2)def swapdim(x, dim1, dim2):
    a = list(range(len(x.shape)))
    a[dim1], a[dim2] = a[dim2], a[dim1]    return x.transpose(a)def drop_path(x, drop_prob = 0., training = False):

    if drop_prob == 0. or not training:        return x
    keep_prob = 1 - drop_prob
    shape = (x.shape[0],) + (1,) * (x.ndim - 1)  
    random_tensor = paddle.to_tensor(keep_prob) + paddle.rand(shape)
    random_tensor = paddle.floor(random_tensor) 
    output = x.divide(keep_prob) * random_tensor    return outputclass DropPath(nn.Layer):

    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 Identity(nn.Layer):                      

    def __init__(self, *args, **kwargs):
        super(Identity, self).__init__() 
    def forward(self, input):
        return input

class Mlp(nn.Layer):
    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0., linear=False):
        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.dwconv = DWConv(hidden_features)
        self.act = act_layer()
        self.fc2 = nn.Linear(hidden_features, out_features)
        self.drop = nn.Dropout(drop)
        self.linear = linear        if self.linear:
            self.relu = nn.ReLU()    def forward(self, x, H, W):
        x = self.fc1(x)        if self.linear:
            x = self.relu(x)
        x = self.dwconv(x, H, W)
        x = self.act(x)
        x = self.drop(x)
        x = self.fc2(x)
        x = self.drop(x)        return xclass Attention(nn.Layer):
    def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0., sr_ratio=1, linear=False):
        super().__init__()        assert dim % num_heads == 0, f"dim {dim} should be divided by num_heads {num_heads}."

        self.dim = dim
        self.num_heads = num_heads
        head_dim = dim // num_heads
        self.scale = qk_scale or head_dim ** -0.5

        self.q = nn.Linear(dim, dim, bias_attr=qkv_bias)
        self.kv = nn.Linear(dim, dim * 2, bias_attr=qkv_bias)
        self.attn_drop = nn.Dropout(attn_drop)
        self.proj = nn.Linear(dim, dim)
        self.proj_drop = nn.Dropout(proj_drop)

        self.linear = linear
        self.sr_ratio = sr_ratio        if not linear:            if sr_ratio > 1:
                self.sr = nn.Conv2D(dim, dim, kernel_size=sr_ratio, stride=sr_ratio)
                self.norm = nn.LayerNorm(dim)        else:
            self.pool = nn.AdaptiveAvgPool2D(7)
            self.sr = nn.Conv2D(dim, dim, kernel_size=1, stride=1)
            self.norm = nn.LayerNorm(dim)
            self.act = nn.GELU()    def forward(self, x, H, W):
        B, N, C = x.shape
        q = self.q(x).reshape([B, N, self.num_heads, C // self.num_heads]).transpose([0, 2, 1, 3])        if not self.linear:            if self.sr_ratio > 1:
                x_ = x.transpose([0, 2, 1]).reshape([B, C, H, W])
                x_ = self.sr(x_).reshape([B, C, -1]).transpose([0, 2, 1])
                x_ = self.norm(x_)
                kv = self.kv(x_).reshape([B, -1, 2, self.num_heads, C // self.num_heads]).transpose([2, 0, 3, 1, 4])            else:
                kv = self.kv(x).reshape([B, -1, 2, self.num_heads, C // self.num_heads]).transpose([2, 0, 3, 1, 4])        else:
            x_ = x.transpose([0, 2, 1]).reshape([B, C, H, W])
            x_ = self.sr(self.pool(x_)).reshape([B, C, -1]).transpose([0, 2, 1])
            x_ = self.norm(x_)
            x_ = self.act(x_)
            kv = self.kv(x_).reshape([B, -1, 2, self.num_heads, C // self.num_heads]).transpose([2, 0, 3, 1, 4])
        k, v = kv[0], kv[1]

        attn = (q @ swapdim(k, -2, -1)) * self.scale
        attn = F.softmax(attn, axis=-1)
        attn = self.attn_drop(attn)

        x = swapdim((attn @ v), 1, 2).reshape([B, N, C])
        x = self.proj(x)
        x = self.proj_drop(x)        return xclass Block(nn.Layer):

    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, sr_ratio=1, linear=False):
        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, sr_ratio=sr_ratio, linear=linear)        # 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 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, linear=linear)    def forward(self, x, H, W):
        x = x + self.drop_path(self.attn(self.norm1(x), H, W))
        x = x + self.drop_path(self.mlp(self.norm2(x), H, W))        return xclass OverlapPatchEmbed(nn.Layer):
    """ Image to Patch Embedding
    """

    def __init__(self, img_size=224, patch_size=7, stride=4, in_chans=3, embed_dim=768):
        super().__init__()
        img_size = to_2tuple(img_size)
        patch_size = to_2tuple(patch_size)

        self.img_size = img_size
        self.patch_size = patch_size
        self.H, self.W = img_size[0] // patch_size[0], img_size[1] // patch_size[1]
        self.num_patches = self.H * self.W
        self.proj = nn.Conv2D(in_chans, embed_dim, kernel_size=patch_size, stride=stride,
                              padding=(patch_size[0] // 2, patch_size[1] // 2))
        self.norm = nn.LayerNorm(embed_dim)    def forward(self, x):
        x = self.proj(x)
        _, _, H, W = x.shape
        x = x.flatten(2)
        x = swapdim(x, 1, 2)
        x = self.norm(x)        return x, H, Wclass PyramidVisionTransformerV2(nn.Layer):
    def __init__(self, img_size=224, patch_size=16, in_chans=3, num_classes=1000, embed_dims=[64, 128, 256, 512],
                 num_heads=[1, 2, 4, 8], mlp_ratios=[4, 4, 4, 4], qkv_bias=False, qk_scale=None, drop_rate=0.,
                 attn_drop_rate=0., drop_path_rate=0., norm_layer=nn.LayerNorm,
                 depths=[3, 4, 6, 3], sr_ratios=[8, 4, 2, 1], num_stages=4, linear=False):
        super().__init__()
        self.num_classes = num_classes
        self.depths = depths
        self.num_stages = num_stages

        dpr = [x for x in paddle.linspace(0, drop_path_rate, sum(depths))]  # stochastic depth decay rule
        cur = 0

        for i in range(num_stages):
            patch_embed = OverlapPatchEmbed(img_size=img_size if i == 0 else img_size // (2 ** (i + 1)),
                                            patch_size=7 if i == 0 else 3,
                                            stride=4 if i == 0 else 2,
                                            in_chans=in_chans if i == 0 else embed_dims[i - 1],
                                            embed_dim=embed_dims[i])

            block = nn.LayerList([Block(
                dim=embed_dims[i], num_heads=num_heads[i], mlp_ratio=mlp_ratios[i], qkv_bias=qkv_bias, qk_scale=qk_scale,
                drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[cur + j], norm_layer=norm_layer,
                sr_ratio=sr_ratios[i], linear=linear)                for j in range(depths[i])])
            norm = norm_layer(embed_dims[i])
            cur += depths[i]            setattr(self, f"patch_embed{i + 1}", patch_embed)            setattr(self, f"block{i + 1}", block)            setattr(self, f"norm{i + 1}", norm)        # classification head
        self.head = nn.Linear(embed_dims[3], num_classes) if num_classes > 0 else Identity()    def freeze_patch_emb(self):
        self.patch_embed1.requires_grad = False

    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 Identity()    def forward_features(self, x):
        B = x.shape[0]        for i in range(self.num_stages):
            patch_embed = getattr(self, f"patch_embed{i + 1}")
            block = getattr(self, f"block{i + 1}")
            norm = getattr(self, f"norm{i + 1}")
            x, H, W = patch_embed(x)            for blk in block:
                x = blk(x, H, W)
            x = norm(x)            if i != self.num_stages - 1:
                x = x.reshape([B, H, W, -1]).transpose([0, 3, 1, 2])        return x.mean(axis=1)    def forward(self, x):
        x = self.forward_features(x)
        x = self.head(x)        return xclass DWConv(nn.Layer):
    def __init__(self, dim=768):
        super(DWConv, self).__init__()
        self.dwconv = nn.Conv2D(dim, dim, 3, 1, 1, bias_attr=True, groups=dim)    def forward(self, x, H, W):
        B, N, C = x.shape
        x = swapdim(x, 1, 2)
        x = x.reshape([B, C, H, W])
        x = self.dwconv(x)
        x = x.flatten(2)
        x = swapdim(x, 1, 2)        return x

def pvt_v2_b0(**kwargs):
    model = PyramidVisionTransformerV2(
        patch_size=4, embed_dims=[32, 64, 160, 256], num_heads=[1, 2, 5, 8], mlp_ratios=[8, 8, 4, 4], qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, epsilon=1e-6), depths=[2, 2, 2, 2], sr_ratios=[8, 4, 2, 1],
        **kwargs)    return modeldef pvt_v2_b1(**kwargs):
    model = PyramidVisionTransformerV2(
        patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[8, 8, 4, 4], qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, epsilon=1e-6), depths=[2, 2, 2, 2], sr_ratios=[8, 4, 2, 1],
        **kwargs)    return modeldef pvt_v2_b2(**kwargs):
    model = PyramidVisionTransformerV2(
        patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[8, 8, 4, 4], qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, epsilon=1e-6), depths=[3, 4, 6, 3], sr_ratios=[8, 4, 2, 1], **kwargs)    return modeldef pvt_v2_b3(**kwargs):
    model = PyramidVisionTransformerV2(
        patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[8, 8, 4, 4], qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, epsilon=1e-6), depths=[3, 4, 18, 3], sr_ratios=[8, 4, 2, 1],
        **kwargs)    return modeldef pvt_v2_b4(**kwargs):
    model = PyramidVisionTransformerV2(
        patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[8, 8, 4, 4], qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, epsilon=1e-6), depths=[3, 8, 27, 3], sr_ratios=[8, 4, 2, 1],
        **kwargs)    return modeldef pvt_v2_b5(**kwargs):
    model = PyramidVisionTransformerV2(
        patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[4, 4, 4, 4], qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, epsilon=1e-6), depths=[3, 6, 40, 3], sr_ratios=[8, 4, 2, 1],
        **kwargs)    return modeldef pvt_v2_b2_li(**kwargs):
    model = PyramidVisionTransformerV2(
        patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[8, 8, 4, 4], qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, epsilon=1e-6), depths=[3, 4, 6, 3], sr_ratios=[8, 4, 2, 1], linear=True, 
        **kwargs)    return model

# 模型各层查看m = pvt_v2_b0()print(m)# 前向计算x = paddle.randn([2, 3, 224, 224])
out = m(x)
loss = out.sum()
loss.backward()print('Single iteration completed successfully')

# pvt v2 b0m = pvt_v2_b0()
m.set_state_dict(paddle.load('/home/aistudio/data/data97429/pvt_v2_b0.pdparams'))# pvt v2 b1m = pvt_v2_b1()
m.set_state_dict(paddle.load('/home/aistudio/data/data97429/pvt_v2_b1.pdparams'))# pvt v2 b2m = pvt_v2_b2()
m.set_state_dict(paddle.load('/home/aistudio/data/data97429/pvt_v2_b2.pdparams'))# pvt v2 b2 linearm = pvt_v2_b2_li()
m.set_state_dict(paddle.load('/home/aistudio/data/data97429/pvt_v2_b2_li.pdparams'))# pvt v2 b3m = pvt_v2_b3()
m.set_state_dict(paddle.load('/home/aistudio/data/data97429/pvt_v2_b3.pdparams'))# pvt v2 b4m = pvt_v2_b4()
m.set_state_dict(paddle.load('/home/aistudio/data/data97429/pvt_v2_b4.pdparams'))# pvt v2 b5m = pvt_v2_b5()
m.set_state_dict(paddle.load('/home/aistudio/data/data97429/pvt_v2_b5.pdparams'))

import paddle.vision.transforms as Tfrom paddle.vision.datasets import Cifar10

paddle.set_device('gpu')#数据准备transform = T.Compose([
    T.Resize(size=(224,224)),
    T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225],data_format='HWC'),
    T.ToTensor()
])

train_dataset = Cifar10(mode='train', transform=transform)
val_dataset = Cifar10(mode='test',  transform=transform)

m = pvt_v2_b0()
m.set_state_dict(paddle.load('/home/aistudio/data/data97429/pvt_v2_b0.pdparams'))
model = paddle.Model(m)

model.prepare(optimizer=paddle.optimizer.AdamW(learning_rate=0.0001, parameters=model.parameters()),
              loss=paddle.nn.CrossEntropyLoss(),
              metrics=paddle.metric.Accuracy())

visualdl=paddle.callbacks.VisualDL(log_dir='visual_log') # 开启训练可视化model.fit(
    train_data=train_dataset, 
    eval_data=val_dataset, 
    batch_size=16, 
    epochs=5,
    verbose=1,
    callbacks=[visualdl] 
)