resnet32

.gitignore

@@ -59,7 +59,7 @@ coverage.xml
 *.pot
 
 # Django stuff:
-*.log
+#*.log
 local_settings.py
 
 # Flask stuff:

README.md

@@ -0,0 +1,10 @@
+# train resnet imagenet with caffe
+### resnet-18
+### resnet-32
+This is a bottleneck architecture,  
+Since there's no strong data augmentation and 10-crop test in caffe, the results maybe a bit low.  
+test accuracy: accuracy@1 = 0.67892, accuracy@5 = 0.88164  
+training loss for resnet-32 is shown below:  
+the pretrained model is provided [here]()
+![a](resnet_32/loss.png)
+

resnet_18/resnet_18.prototxt

@@ -56,7 +56,7 @@ layer {
     }
     bias_filler {
       type: "constant"
-      value: 0.2
+      value: 0
     }
   }
 }
@@ -108,10 +108,6 @@ layer {
     weight_filler {
       type: "msra"
     }
-    bias_filler {
-      type: "constant"
-      value: 0.2
-    }
   }
 }
 layer {
@@ -145,10 +141,6 @@ layer {
     weight_filler {
       type: "msra"
     }
-    bias_filler {
-      type: "constant"
-      value: 0.2
-    }
   }
 }
 layer {
@@ -188,10 +180,6 @@ layer {
     weight_filler {
       type: "msra"
     }
-    bias_filler {
-      type: "constant"
-      value: 0.2
-    }
   }
 }
 layer {
@@ -231,10 +219,6 @@ layer {
     weight_filler {
       type: "msra"
     }
-    bias_filler {
-      type: "constant"
-      value: 0.2
-    }
   }
 }
 layer {
@@ -281,10 +265,6 @@ layer {
     weight_filler {
       type: "msra"
     }
-    bias_filler {
-      type: "constant"
-      value: 0.2
-    }
   }
 }
 layer {
@@ -324,10 +304,6 @@ layer {
     weight_filler {
       type: "msra"
     }
-    bias_filler {
-      type: "constant"
-      value: 0.2
-    }
   }
 }
 layer {
@@ -367,10 +343,6 @@ layer {
     weight_filler {
       type: "msra"
     }
-    bias_filler {
-      type: "constant"
-      value: 0.2
-    }
   }
 }
 layer {
@@ -417,10 +389,6 @@ layer {
     weight_filler {
       type: "msra"
     }
-    bias_filler {
-      type: "constant"
-      value: 0.2
-    }
   }
 }
 layer {
@@ -460,10 +428,6 @@ layer {
     weight_filler {
       type: "msra"
     }
-    bias_filler {
-      type: "constant"
-      value: 0.2
-    }
   }
 }
 layer {
@@ -503,10 +467,6 @@ layer {
     weight_filler {
       type: "msra"
     }
-    bias_filler {
-      type: "constant"
-      value: 0.2
-    }
   }
 }
 layer {
@@ -553,10 +513,6 @@ layer {
     weight_filler {
       type: "msra"
     }
-    bias_filler {
-      type: "constant"
-      value: 0.2
-    }
   }
 }
 layer {
@@ -590,10 +546,6 @@ layer {
     weight_filler {
       type: "msra"
     }
-    bias_filler {
-      type: "constant"
-      value: 0.2
-    }
   }
 }
 layer {
@@ -633,10 +585,6 @@ layer {
     weight_filler {
       type: "msra"
     }
-    bias_filler {
-      type: "constant"
-      value: 0.2
-    }
   }
 }
 layer {
@@ -676,10 +624,6 @@ layer {
     weight_filler {
       type: "msra"
     }
-    bias_filler {
-      type: "constant"
-      value: 0.2
-    }
   }
 }
 layer {
@@ -726,10 +670,6 @@ layer {
     weight_filler {
       type: "msra"
     }
-    bias_filler {
-      type: "constant"
-      value: 0.2
-    }
   }
 }
 layer {
@@ -769,10 +709,6 @@ layer {
     weight_filler {
       type: "msra"
     }
-    bias_filler {
-      type: "constant"
-      value: 0.2
-    }
   }
 }
 layer {
@@ -812,10 +748,6 @@ layer {
     weight_filler {
       type: "msra"
     }
-    bias_filler {
-      type: "constant"
-      value: 0.2
-    }
   }
 }
 layer {