【Hackathon 5th No.69】 分类大模型--人体视觉任务SOLIDER

docs/zh_CN/models/LUPerson/solider.md

@@ -0,0 +1,27 @@
+# Solider
+
+-----
+## 目录
+
+- [1. 模型介绍](#1)
+- [2. 对齐日志、模型](#2)
+
+<a name='1'></a>
+
+## 1. 模型介绍
+
+Solider是一个语义可控的自监督学习框架，可以从大量未标记的人体图像中学习一般的人类表征，从而最大限度地有利于下游以人类为中心的任务。与已有的自监督学习方法不同，该方法利用人体图像中的先验知识建立伪语义标签，并将更多的语义信息引入到学习的表示中。同时，不同的下游任务往往需要不同比例的语义信息和外观信息，单一的学习表示不能满足所有需求。为了解决这一问题，Solider引入了一种带有语义控制器的条件网络，可以满足下游任务的不同需求。[论文地址](https://arxiv.org/abs/2303.17602)。
+
+<a name='2'></a>
+
+## 2. 对齐日志、模型
+
+| model                         | weight                                                       | log                                                          |
+| ----------------------------- | ------------------------------------------------------------ | ------------------------------------------------------------ |
+| swin_tiny_patch4_window7_224  | https://paddleclas.bj.bcebos.com/models/SOILDER/SwinTransformer_tiny_patch4_window7_224_pretrained.pdparams | 链接：https://pan.baidu.com/s/1W5zUFboMMhXETy4HEWbM3Q?pwd=45nx <br/>提取码：45nx |
+| swin_small_patch4_window7_224 | https://paddleclas.bj.bcebos.com/models/SOILDER/SwinTransformer_small_patch4_window7_224_pretrained.pdparams | 链接：https://pan.baidu.com/s/1sqcUdfv6FyhW9_QgxBUPWA?pwd=letv <br/>提取码：letv |
+| swin_base_patch4_window7_224  | https://paddleclas.bj.bcebos.com/models/SOILDER/SwinTransformer_base_patch4_window7_224_pretrained.pdparams | 链接：https://pan.baidu.com/s/1S2TgDxDRa72C_3FrP8duiA?pwd=u3d2 <br/>提取码：u3d2 |
+
+[1]：基于  LUPerson 数据集预训练
+
+<a name='3'></a>

ppcls/arch/backbone/__init__.py

@@ -91,6 +91,9 @@
 from .variant_models.pp_lcnetv2_variant import PPLCNetV2_base_ShiTu
 from .variant_models.efficientnet_variant import EfficientNetB3_watermark
 from .variant_models.foundation_vit_variant import CLIP_large_patch14_224_aesthetic
+from .variant_models.swin_transformer_variant import SwinTransformer_tiny_patch4_window7_224_SOLIDER
+from .variant_models.swin_transformer_variant import SwinTransformer_small_patch4_window7_224_SOLIDER
+from .variant_models.swin_transformer_variant import SwinTransformer_base_patch4_window7_224_SOLIDER
 from .model_zoo.adaface_ir_net import AdaFace_IR_18, AdaFace_IR_34, AdaFace_IR_50, AdaFace_IR_101, AdaFace_IR_152, AdaFace_IR_SE_50, AdaFace_IR_SE_101, AdaFace_IR_SE_152, AdaFace_IR_SE_200
 from .model_zoo.wideresnet import WideResNet
 from .model_zoo.uniformer import UniFormer_small, UniFormer_small_plus, UniFormer_small_plus_dim64, UniFormer_base, UniFormer_base_ls

ppcls/arch/backbone/legendary_models/swin_transformer.py

@@ -359,12 +359,8 @@ def __init__(self,
         self.window_size = window_size
         self.shift_size = shift_size
         self.mlp_ratio = mlp_ratio
-        if min(self.input_resolution) <= self.window_size:
-            # if window size is larger than input resolution, we don't partition windows
-            self.shift_size = 0
-            self.window_size = min(self.input_resolution)
-        assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size"
 
+        self.check_condition()
         self.norm1 = norm_layer(dim)
         self.attn = WindowAttention(
             dim,
@@ -412,6 +408,13 @@ def __init__(self,
 
         self.register_buffer("attn_mask", attn_mask)
 
+    def check_condition(self):
+        if min(self.input_resolution) <= self.window_size:
+            # if window size is larger than input resolution, we don't partition windows
+            self.shift_size = 0
+            self.window_size = min(self.input_resolution)
+        assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size"
+
     def forward(self, x):
         H, W = self.input_resolution
         B, L, C = x.shape
@@ -835,7 +838,8 @@ def _load_pretrained(pretrained,
                      model_url,
                      use_ssld=False,
                      use_imagenet22k_pretrained=False,
-                     use_imagenet22kto1k_pretrained=False):
+                     use_imagenet22kto1k_pretrained=False,
+                     **kwargs):
     if pretrained is False:
         pass
     elif pretrained is True:
@@ -988,4 +992,4 @@ def SwinTransformer_large_patch4_window12_384(
         use_ssld=use_ssld,
         use_imagenet22k_pretrained=use_imagenet22k_pretrained,
         use_imagenet22kto1k_pretrained=use_imagenet22kto1k_pretrained)
-    return model
+    return model

ppcls/arch/backbone/variant_models/__init__.py

@@ -2,3 +2,6 @@
 from .vgg_variant import VGG19Sigmoid
 from .pp_lcnet_variant import PPLCNet_x2_5_Tanh
 from .pp_lcnetv2_variant import PPLCNetV2_base_ShiTu
+from .swin_transformer_variant import SwinTransformer_base_patch4_window7_224_SOLIDER
+from .swin_transformer_variant import SwinTransformer_small_patch4_window7_224_SOLIDER
+from .swin_transformer_variant import SwinTransformer_tiny_patch4_window7_224_SOLIDER