本文复现了PVT v2模型，其基于v1改进，亮点是Linear SRA。代码包含导入包、基础模块定义、模型组网等部分，还提供了不同缩放结构及预训练权重。通过在Cifar10数据集上训练5轮验证性能，模型表现良好。PVT v2引入卷积等操作提升性能，参数量和计算量较小，下游任务表现佳。

前言

Hi guy，我们怎么又见面了，(俗套的开场白)，哈哈哈哈，那么这次来复现一个PVT v2，它是基于v1进行改动

完整代码

导入所需要的包

In [?]

import paddleimport paddle.nn as nnimport paddle.nn.functional as Ffrom functools import partialimport mathtrunc_normal_ = nn.initializer.TruncatedNormal(std=.02)zeros_ = nn.initializer.Constant(value=0.)ones_ = nn.initializer.Constant(value=1.)kaiming_normal_ = nn.initializer.KaimingNormal()登录后复制

/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/layers/utils.py:26: DeprecationWarning: `np.int` is a deprecated alias for the builtin `int`. To silence this warning, use `int` by itself. Doing this will not modify any behavior and is safe. When replacing `np.int`, you may wish to use e.g. `np.int64` or `np.int32` to specify the precision. If you wish to review your current use, check the release note link for additional information.Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations def convert_to_list(value, n, name, dtype=np.int):登录后复制

基础模块定义

In [?]

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登录后复制

模型组网

网络大概结构如下图所示

其中亮点是 PVT v2 的 Linear SRA

In [?]

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登录后复制

模型缩放

模型各种缩放结构官方性能如下所示

In [?]

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登录后复制

查看模型

In [?]

# 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')登录后复制

预训练权重加载

Results on ImageNet-1K

Model# ParamTop-1 Acc.Top-5 Acc.pvt v2 b03M0.7030.900pvt v2 b114M0.7840.943pvt v2 b225M0.8170.959pvt v2 b2 li45M0.8290.964pvt v2 b362M0.8340.967pvt v2 b482M0.8350.966pvt v2 b522M0.8190.960In [?]

# 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'))登录后复制

Cifar10 验证性能

数据准备

In [15]

import paddle.vision.transforms as Tfrom paddle.vision.datasets import Cifar10paddle.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)登录后复制

Cache file /home/aistudio/.cache/paddle/dataset/cifar/cifar-10-python.tar.gz not found, downloading https://dataset.bj.bcebos.com/cifar/cifar-10-python.tar.gz Begin to downloadDownload finished登录后复制

模型准备

In [16]

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

开始训练

由于时间篇幅只训练5轮，感兴趣的同学可以继续训练

In [17]

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] )登录后复制

The loss value printed in the log is the current step, and the metric is the average value of previous step.Epoch 1/5登录后复制

/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/layers/utils.py:77: DeprecationWarning: Using or importing the ABCs from 'collections' instead of from 'collections.abc' is deprecated, and in 3.8 it will stop working return (isinstance(seq, collections.Sequence) and登录后复制

step 3125/3125 [==============================] - loss: 0.2681 - acc: 0.8151 - 124ms/step Eval begin...The loss value printed in the log is the current batch, and the metric is the average value of previous step.step 625/625 [==============================] - loss: 0.2054 - acc: 0.9154 - 57ms/step Eval samples: 10000Epoch 2/5step 3125/3125 [==============================] - loss: 0.3357 - acc: 0.9383 - 126ms/step Eval begin...The loss value printed in the log is the current batch, and the metric is the average value of previous step.step 625/625 [==============================] - loss: 0.1274 - acc: 0.9245 - 58ms/step Eval samples: 10000Epoch 3/5step 3125/3125 [==============================] - loss: 0.0413 - acc: 0.9594 - 126ms/step Eval begin...The loss value printed in the log is the current batch, and the metric is the average value of previous step.step 625/625 [==============================] - loss: 0.0439 - acc: 0.9367 - 57ms/step Eval samples: 10000Epoch 4/5step 3125/3125 [==============================] - loss: 0.0458 - acc: 0.9696 - 126ms/step Eval begin...The loss value printed in the log is the current batch, and the metric is the average value of previous step.step 625/625 [==============================] - loss: 0.2247 - acc: 0.9351 - 56ms/step Eval samples: 10000Epoch 5/5step 3125/3125 [==============================] - loss: 0.0024 - acc: 0.9760 - 128ms/step Eval begin...The loss value printed in the log is the current batch, and the metric is the average value of previous step.step 625/625 [==============================] - loss: 0.0181 - acc: 0.9374 - 57ms/step Eval samples: 10000登录后复制

训练可视化

总结

• PVT v2 引入了卷积操作、zero-padding、avgpool的注意力层，从三个方面提升了性能
• 相比同时期的ViT模型，具有更小的参数量和计算量
• 在下游任务，PVT v2 展现了良好的性能