update to new result

README.md

@@ -1,7 +1,7 @@
 # Deep Residual Net
 Example code for [Deep Residual Learning for Image Recognition](http://arxiv.org/abs/1512.03385)  
-* Run this script by ```python resnet-small.py``` for 70 epochs get a train accuracy around 84% and validation accuracy around 93%  
-* Then change the learning rate to 0.01, running this training from 70th epoch for 40 iterations, and get a train accuracy around 99% and test accuracy around 89%  
+* Run this script by ```python resnet-small.py``` for 100 epochs get a train accuracy around 89.47% and validation accuracy around 85.95%  
+* Then change the learning rate to 0.01, running this training from 70th epoch for 50 iterations, and get a train accuracy around 98.72% and test accuracy around 89.77%  
 
 ## Differences to the Paper
 * 1*1 convolution operators are used for increasing dimensions.

resnet-small.py

@@ -93,7 +93,7 @@ def get_dataiter(batch_size=128):
     # g.view()
     batch_size = 128
     train_dataiter, test_dataiter = get_dataiter(batch_size=batch_size)
-    finetune=False
+    finetune=True
     if finetune==False:
         model = mx.model.FeedForward(ctx=mx.gpu(0), symbol=softmax, num_epoch=70, learning_rate=0.1, momentum=0.9, wd=0.0001, \
                                  initializer=mx.init.Xavier(rnd_type='gaussian', factor_type="in", magnitude=2),
@@ -103,6 +103,6 @@ def get_dataiter(batch_size=128):
                                  )
         model.fit(X=train_dataiter, eval_data=test_dataiter, batch_end_callback=mx.callback.Speedometer(batch_size),epoch_end_callback=mx.callback.do_checkpoint("./models/resnet"))
     else:
-        loaded = mx.model.FeedForward.load('models/resnet', 70)
+        loaded = mx.model.FeedForward.load('models/resnet', 100)
         continue_model = mx.model.FeedForward(ctx=mx.gpu(0), symbol = loaded.symbol, arg_params = loaded.arg_params, aux_params = loaded.aux_params, num_epoch=10000, learning_rate=0.01, momentum=0.9, wd=0.0001)
         continue_model.fit(X=train_dataiter, eval_data=test_dataiter, batch_end_callback=mx.callback.Speedometer(batch_size),epoch_end_callback=mx.callback.do_checkpoint("./models/resnet"))