Merge pull request #139 from Gaffey/snn_mlp_branch

Snn mlp branch

hubconf.py

@@ -1,11 +1,15 @@
-# 2020.06.09-GhostNet definition for pytorch hub
+# 2022.09.16-GhostNet & SNN-MLP definition for pytorch hub
 #            Huawei Technologies Co., Ltd. <[email protected]>
 dependencies = ['torch']
 import torch
 from ghostnet_pytorch.ghostnet import ghostnet
+from snnmlp_pytorch.models.snn_mlp import SNNMLP
 
 
 state_dict_url = 'https://github.com/huawei-noah/ghostnet/raw/master/ghostnet_pytorch/models/state_dict_73.98.pth'
+state_dict_url_snnmlp_t = 'https://github.com/huawei-noah/Efficient-AI-Backbones/releases/download/snnmlp/snnmlp_tiny_81.88.pt'
+state_dict_url_snnmlp_s = 'https://github.com/huawei-noah/Efficient-AI-Backbones/releases/download/snnmlp/snnmlp_small_83.30.pt'
+state_dict_url_snnmlp_b = 'https://github.com/huawei-noah/Efficient-AI-Backbones/releases/download/snnmlp/snnmlp_base_83.59.pt'
 
 
 def ghostnet_1x(pretrained=False, **kwargs):
@@ -18,3 +22,36 @@ def ghostnet_1x(pretrained=False, **kwargs):
 	  	  state_dict = torch.hub.load_state_dict_from_url(state_dict_url, progress=True)
 	  	  model.load_state_dict(state_dict)
 	  return model
+
+def snnmlp_t(pretrained=False, **kwargs):
+	  """ # This docstring shows up in hub.help()
+    SNN-MLP tiny model
+    pretrained (bool): kwargs, load pretrained weights into the model
+    """
+	  model = SNNMLP(num_classes=1000, embed_dim=96, depths=[2, 2, 6, 2], drop_path_rate=0.2)
+	  if pretrained:
+	  	  state_dict = torch.hub.load_state_dict_from_url(state_dict_url_snnmlp_t, progress=True)
+	  	  model.load_state_dict(state_dict)
+	  return model
+
+def snnmlp_s(pretrained=False, **kwargs):
+	  """ # This docstring shows up in hub.help()
+    SNN-MLP small model
+    pretrained (bool): kwargs, load pretrained weights into the model
+    """
+	  model = SNNMLP(num_classes=1000, embed_dim=96, depths=[2, 2, 18, 2], drop_path_rate=0.3)
+	  if pretrained:
+	  	  state_dict = torch.hub.load_state_dict_from_url(state_dict_url_snnmlp_s, progress=True)
+	  	  model.load_state_dict(state_dict)
+	  return model
+
+def snnmlp_b(pretrained=False, **kwargs):
+	  """ # This docstring shows up in hub.help()
+    SNN-MLP base model
+    pretrained (bool): kwargs, load pretrained weights into the model
+    """
+	  model = SNNMLP(num_classes=1000, embed_dim=128, depths=[2, 2, 18, 2], drop_path_rate=0.5)
+	  if pretrained:
+	  	  state_dict = torch.hub.load_state_dict_from_url(state_dict_url_snnmlp_b, progress=True)
+	  	  model.load_state_dict(state_dict)
+	  return model

snnmlp_pytorch/readme.md

@@ -45,13 +45,11 @@ Dataset used: [ImageNet2012](http://www.image-net.org/)
 
 ```
 
-Download checkpoints: [[BaiduDisk]](https://pan.baidu.com/s/1YuxSJNOUyPZUKUPy419HPg), password: 02tb.
-
 ## Result
 
-|Model|Params (M)|FLOPs (B)|Top-1|
-|-|-|-|-|
-|SNN-MLP-T|28.3|4.4|81.9|
-|SNN-MLP-S|49.7|8.5|83.3|
-|SNN-MLP-B|87.9|15.2|83.6|
+|Model|Params (M)|FLOPs (B)|Top-1|Download URL|
+|-|-|-|-|-|
+|SNN-MLP-T|28.3|4.4|81.9|[[ckpt]](https://github.com/huawei-noah/Efficient-AI-Backbones/releases/download/snnmlp/snnmlp_tiny_81.88.pt) & [[log]]( https://github.com/huawei-noah/Efficient-AI-Backbones/releases/download/snnmlp/snnmlp_tiny_81.88.log)|
+|SNN-MLP-S|49.7|8.5|83.3|[[ckpt]](https://github.com/huawei-noah/Efficient-AI-Backbones/releases/download/snnmlp/snnmlp_small_83.30.pt) & [[log]]( https://github.com/huawei-noah/Efficient-AI-Backbones/releases/download/snnmlp/snnmlp_small_83.30.log)|
+|SNN-MLP-B|87.9|15.2|83.6|[[ckpt]](https://github.com/huawei-noah/Efficient-AI-Backbones/releases/download/snnmlp/snnmlp_base_83.59.pt) & [[log]]( https://github.com/huawei-noah/Efficient-AI-Backbones/releases/download/snnmlp/snnmlp_base_83.59.log)|