Add UnsuperLearnFindBestWeight to find best from 4000

FilterEvaluator/Evaluator copy.py

@@ -0,0 +1,113 @@
+from __future__ import print_function
+import os
+import sys
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torch.optim as optim
+import torchvision
+from torchvision import datasets, transforms
+import torchvision.models as models
+import matplotlib.pyplot as plt
+import numpy as np
+from torch.utils.data import Dataset, DataLoader
+from PIL import Image
+import random
+import cv2
+
+
+
+CurrentPath = os.path.split(os.path.realpath(__file__))[0]+"/"
+print("Current Path :" + CurrentPath)
+
+sys.path.append(CurrentPath+'../tools')
+sys.path.append(CurrentPath+'../')
+
+import Model as Model
+from tools import utils, Train, Loader
+import EvaluatorUnsuper
+
+
+
+batchsize = 128
+
+
+# model = utils.SetDevice(Model.Net5Grad35())
+# model = utils.SetDevice(Model.Net31535())
+model = utils.SetDevice(Model.Net3Grad335())
+# model = utils.SetDevice(Model.Net3())
+
+
+layers = model.PrintLayer()
+layer = 0
+
+model = utils.LoadModel(model, CurrentPath+"/checkpoint.pkl")
+
+weight = np.load(CurrentPath+"WeightSearch.npy")
+
+
+
+
+
+# traindata, testdata = Loader.MNIST(batchsize)
+# traindata, testdata = Loader.RandomMnist(batchsize, style="Vertical")
+# traindata, testdata = Loader.RandomMnist(batchsize, style="Horizontal")
+# traindata, testdata = Loader.RandomMnist(batchsize, style="VerticalOneLine")
+# traindata, testdata = Loader.RandomMnist(batchsize, style="VerticalZebra")
+# traindata, testdata = Loader.Cifar10Mono(batchsize)
+traindata, testdata = Loader.Cifar10Mono(batchsize, num_workers=0, shuffle=True)
+
+
+
+
+# weight = EvaluatorUnsuper.UnsuperLearnSearchWeight(model, layer, traindata, NumSearch=100000, SearchChannelRatio=8, Interation=10)
+# np.save("WeightSearch.npy", weight)
+
+# weight = EvaluatorUnsuper.UnsuperLearnTrainWeight(model, layer, traindata)
+# np.save("WeightTrain.npy", weight)
+
+
+
+
+weight = np.load(CurrentPath+"WeightSearch.npy")
+
+
+def DistanceOfKernel(k1,k2):
+    k1 = k1.reshape((1,-1))
+    k1r = 1.0/k1.reshape((-1,1))
+    k1dot = np.dot(k1r,k1)
+    k2 = k2.reshape((1,-1))
+    k2r = 1.0/k2.reshape((-1,1))
+    k2dot = np.dot(k2r,k2)
+    diff = np.abs(np.mean(k1dot - k2dot))
+    return diff
+
+
+indexs = np.random.randint(0,weight.shape[0],(10000,2))
+
+mindis = 0
+manindex = []
+for i in indexs:
+    if i[0] == i[1]:
+      continue
+    dis = DistanceOfKernel(weight[i[0]], weight[i[1]])
+    if dis > mindis: 
+      manindex = i
+      mindis = dis
+
+a = weight[manindex[0]]
+b = weight[manindex[1]]
+
+utils.NumpyToImage(a, CurrentPath+"image","a")
+utils.NumpyToImage(b, CurrentPath+"image","b")
+
+a = 0
+
+
+
+
+
+
+# utils.NumpyToImage(weight, CurrentPath+"image")
+# utils.SaveModel(model,CurrentPath+"/checkpoint.pkl")
+print("save model sucess")

FilterEvaluator/Evaluator.py

@@ -1,10 +1,6 @@
 from __future__ import print_function
 import os
 import sys
-
-# import multiprocessing
-# multiprocessing.set_start_method('spawn', True)
-
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
@@ -20,6 +16,7 @@ import random
 import cv2
 
 
+
 CurrentPath = os.path.split(os.path.realpath(__file__))[0]+"/"
 print("Current Path :" + CurrentPath)
 
@@ -28,6 +25,9 @@ sys.path.append(CurrentPath+'../')
 
 import Model as Model
 from tools import utils, Train, Loader
+import EvaluatorUnsuper
+
+
 
 batchsize = 128
 
@@ -40,8 +40,7 @@ model = utils.SetDevice(Model.Net3Grad335())
 
 layers = model.PrintLayer()
 layer = 0
-
-model = utils.LoadModel(model, CurrentPath+"/checkpoint.pkl")
+# model = utils.LoadModel(model, CurrentPath+"/checkpoint.pkl")
 
 
 
@@ -53,132 +52,43 @@ model = utils.LoadModel(model, CurrentPath+"/checkpoint.pkl")
 # traindata, testdata = Loader.RandomMnist(batchsize, style="VerticalOneLine")
 # traindata, testdata = Loader.RandomMnist(batchsize, style="VerticalZebra")
 # traindata, testdata = Loader.Cifar10Mono(batchsize)
-traindata, testdata = Loader.Cifar10Mono(batchsize, num_workers=2, shuffle=True)
-
-
-
-def GetScore(netmodel,layer,SearchLayer,DataSet,Interation=-1):
-    netmodel.eval()
-    sample = utils.SetDevice(torch.empty((SearchLayer.out_channels,0)))
-
-    layer = layer-1
-
-    # layerout = []
-    #     layerint = []
-    #     def getnet(self, input, output):
-    #         layerout.append(output)
-    #         layerint.append(input)
-    # handle = netmodel.features[layer].register_forward_hook(getnet)
-    # netmodel.ForwardLayer(data,layer)
-    # output = layerout[0][:,:,:,:]
-    # handle.remove()
-
-
-    for batch_idx, (data, target) in enumerate(DataSet):
-        data = utils.SetDevice(data)
-        target = utils.SetDevice(target)
-        output = netmodel.ForwardLayer(data,layer)
-        output = SearchLayer(output)
-        data.detach()
-        target.detach()
-        
-        output = torch.reshape(output.transpose(0,1),(SearchLayer.out_channels,-1))
-        sample = torch.cat((sample,output),1)
-        if Interation > 0 and batch_idx >= (Interation-1):
-            break
-
-    sample_mean=torch.mean(sample,dim=1,keepdim=True)
-    dat1 = torch.mean(torch.abs(sample - sample_mean),dim=1,keepdim=True)
-    dat2 = (sample - sample_mean)/dat1
-    dat2 = torch.mean(dat2 * dat2,dim=1)
-    return dat2.cpu().detach().numpy()
-
-def UnsuperLearnSearchWeight(model, layer, dataloader, NumSearch=10000, SaveChannelRatio=500, SearchChannelRatio=1, Interation=10):
-    tl = model.features[layer]
-    newlayer = nn.Conv2d(tl.in_channels, tl.out_channels * SearchChannelRatio, tl.kernel_size,
-                         tl.stride, tl.padding, tl.dilation, tl.groups, tl.bias, tl.padding_mode)
-    newlayer = utils.SetDevice(newlayer)
-
-    newchannels = tl.out_channels * SaveChannelRatio
-    newweightshape = list(newlayer.weight.data.shape)
-    
-    minactive = np.empty((0))
-    minweight = np.empty([0,newweightshape[1],newweightshape[2],newweightshape[3]])
-
-    for i in range(NumSearch):
-        newweight = np.random.uniform(-1.0,1.0,newweightshape).astype("float32")
-        newlayer.weight.data=utils.SetDevice(torch.from_numpy(newweight))
-
-        score = GetScore(model, layer, newlayer, dataloader, Interation)
-
-        minactive = np.append(minactive, score)
-        minweight = np.concatenate((minweight, newweight))
-
-        index = minactive.argsort()
-        minactive = minactive[index[0:newchannels]]
-        minweight = minweight[index[0:newchannels]]
-        print("search random :" + str(i))
-        if i % (NumSearch/10) == 0:
-            tl.data=utils.SetDevice(torch.from_numpy(minweight[0:tl.out_channels]))
-            utils.SaveModel(model, CurrentPath+"/checkpoint.pkl")
-    
-    tl.data=utils.SetDevice(torch.from_numpy(minweight[0:tl.out_channels]))
-    return minweight
-
-
-def TrainLayer(netmodel, layer, SearchLayer, DataSet, Epoch=100):
-    netmodel.eval()
-    sample = utils.SetDevice(torch.empty((SearchLayer.out_channels,0)))
-    layer = layer-1
-    SearchLayer.weight.data.requires_grad=True
-    optimizer = optim.SGD(SearchLayer.parameters(), lr=0.1)
-    for e in range(Epoch):
-        for batch_idx, (data, target) in enumerate(DataSet):
-            optimizer.zero_grad()
-            data = utils.SetDevice(data)
-            target = utils.SetDevice(target)
-            output = netmodel.ForwardLayer(data,layer)
-            output = SearchLayer(output)
-            sample = torch.reshape(output.transpose(0,1),(SearchLayer.out_channels,-1))
-            sample_mean=torch.mean(sample,dim=1,keepdim=True)
-            dat1 = torch.mean(torch.abs(sample - sample_mean),dim=1,keepdim=True)
-            dat2 = (sample - sample_mean)/dat1
-            dat2 = torch.mean(dat2 * dat2,dim=1)
-            label = dat2*0.5
-            loss = F.l1_loss(dat2, label)
-            loss.backward()
-            optimizer.step()
-        print("  epoch :" + str(e))
-    return SearchLayer.weight.data.cpu().detach().numpy()
-
-def UnsuperLearnTrainWeight(model, layer, dataloader, NumTrain=500, TrainChannelRatio=1, Epoch=20):
-    tl = model.features[layer]
-    newlayer = nn.Conv2d(tl.in_channels, tl.out_channels * TrainChannelRatio, tl.kernel_size,
-                         tl.stride, tl.padding, tl.dilation, tl.groups, tl.bias, tl.padding_mode)
-    newlayer = utils.SetDevice(newlayer)
-    newweightshape = list(newlayer.weight.data.shape)
-    minactive = np.empty((0))
-    minweight = np.empty([0,newweightshape[1],newweightshape[2],newweightshape[3]])
-    for i in range(NumTrain):
-        newweight = np.random.uniform(-1.0,1.0,newweightshape).astype("float32")
-        newlayer.weight.data=utils.SetDevice(torch.from_numpy(newweight))
-        newweight = TrainLayer(model, layer, newlayer, dataloader, Epoch)
-        minweight = np.concatenate((minweight, newweight))
-        print("search :" + str(i))
-    return minweight
+traindata, testdata = Loader.Cifar10Mono(batchsize, num_workers=0, shuffle=True)
 
 
 
 
-
-# weight = UnsuperLearnSearchWeight(model, layer, traindata, NumSearch=100000, SearchChannelRatio=8, Interation=10)
+# weight = EvaluatorUnsuper.UnsuperLearnSearchWeight(model, layer, traindata, NumSearch=100000, SearchChannelRatio=8, Interation=10)
 # np.save("WeightSearch.npy", weight)
 
-weight = UnsuperLearnTrainWeight(model, layer, traindata)
-np.save("WeightTrain.npy", weight)
+# weight = EvaluatorUnsuper.UnsuperLearnTrainWeight(model, layer, traindata)
+# np.save("WeightTrain.npy", weight)
 
 
 
-utils.NumpyToImage(weight, CurrentPath+"image")
+
+# weight = np.load(CurrentPath+"WeightSearch.npy")
+# bestweight = EvaluatorUnsuper.UnsuperLearnFindBestWeight(model,layer,weight,traindata,128,100000)
+# np.save(CurrentPath+"bestweightSearch.npy", bestweight)
+# utils.NumpyToImage(bestweight, CurrentPath+"image")
+
+
+# weight = np.load(CurrentPath+"WeightTrain.npy")
+# bestweight = EvaluatorUnsuper.UnsuperLearnFindBestWeight(model,layer,weight,traindata,128,100000)
+# np.save(CurrentPath+"bestweightTrain.npy", bestweight)
+# utils.NumpyToImage(bestweight, CurrentPath+"image")
+
+
+
+weight = np.load(CurrentPath+"bestweightSearch.npy")
+EvaluatorUnsuper.SetModelConvWeight(model,layer,weight)
+utils.SaveModel(model,CurrentPath+"/checkpointSearch.pkl")
+
+weight = np.load(CurrentPath+"bestweightTrain.npy")
+EvaluatorUnsuper.SetModelConvWeight(model,layer,weight)
+utils.SaveModel(model,CurrentPath+"/checkpointTrain.pkl")
+
+
+
+# utils.NumpyToImage(weight, CurrentPath+"image")
 # utils.SaveModel(model,CurrentPath+"/checkpoint.pkl")
 print("save model sucess")

FilterEvaluator/EvaluatorUnsuper.py

@@ -1,10 +1,6 @@
 from __future__ import print_function
 import os
 import sys
-
-# import multiprocessing
-# multiprocessing.set_start_method('spawn', True)
-
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
@@ -19,50 +15,13 @@ from PIL import Image
 import random
 import cv2
 
-
 CurrentPath = os.path.split(os.path.realpath(__file__))[0]+"/"
-print("Current Path :" + CurrentPath)
-
 sys.path.append(CurrentPath+'../tools')
 sys.path.append(CurrentPath+'../')
 
 import Model as Model
 from tools import utils, Train, Loader
 
-batchsize = 128
-
-
-# model = utils.SetDevice(Model.Net5Grad35())
-# model = utils.SetDevice(Model.Net31535())
-model = utils.SetDevice(Model.Net3Grad335())
-# model = utils.SetDevice(Model.Net3())
-
-
-layers = model.PrintLayer()
-layer = 0
-
-model = utils.LoadModel(model, CurrentPath+"/checkpoint.pkl")
-
-
-
-
-
-
-# traindata, testdata = Loader.MNIST(batchsize)
-# traindata, testdata = Loader.RandomMnist(batchsize, style="Vertical")
-# traindata, testdata = Loader.RandomMnist(batchsize, style="Horizontal")
-# traindata, testdata = Loader.RandomMnist(batchsize, style="VerticalOneLine")
-# traindata, testdata = Loader.RandomMnist(batchsize, style="VerticalZebra")
-# traindata, testdata = Loader.Cifar10Mono(batchsize)
-traindata, testdata = Loader.Cifar10Mono(batchsize, num_workers=2, shuffle=True)
-
-
-
-
-
-
-
-
 
 def GetScore(netmodel,layer,SearchLayer,DataSet,Interation=-1):
     netmodel.eval()
@@ -175,35 +134,49 @@ def UnsuperLearnTrainWeight(model, layer, dataloader, NumTrain=500, TrainChannel
     return minweight
 
 
+def UnsuperLearnFindBestWeight(netmodel, layer, weight, dataloader, databatchs=128,interation=10000):
+    weight = weight.astype("float32")
+    netmodel.eval()
+    tl = netmodel.features[layer]
+    outchannels = tl.out_channels
+    newlayer = nn.Conv2d(tl.in_channels, tl.out_channels, tl.kernel_size,
+                         tl.stride, tl.padding, tl.dilation, tl.groups, tl.bias, tl.padding_mode)
+    newlayer = utils.SetDevice(newlayer)
+
+    datas = []
+    for batch_idx, (data, target) in enumerate(dataloader):
+        datas.append(utils.SetDevice(data))
+        if batch_idx >= databatchs-1:
+            break
+    indexs = np.random.randint(0,weight.shape[0],interation*outchannels).reshape(interation,-1)
+    entropys = []
+    forwardlayer = layer -1
+    shift = []
+    for i in range(outchannels):
+        shift.append(1<<i)
+    shift = utils.SetDevice(torch.from_numpy(np.array(shift).astype("uint8")))
+    for i in range(len(indexs)):
+        newlayer.weight.data=utils.SetDevice(torch.from_numpy(weight[indexs[i]]))
+        outputs = []
+        for d in datas:
+            outputs.append(newlayer(netmodel.ForwardLayer(d,forwardlayer)))
+        outputs = torch.cat(outputs).transpose(0,1)
+        outputs = outputs.reshape(outputs.shape[0],-1)
+        mean = outputs.mean(1)
+        outputs = outputs.transpose(0,1)
+        posi = outputs > mean
+        posi = posi*shift
+        sums = posi.sum(1).type(torch.float32)
+        histc = sums.histc(256,0,255).type(torch.float32)
+        histc = histc[histc>0]
+        histc = histc/histc.sum()
+        entropys.append((histc.log2()*histc).sum())
+        print("search index : " + str(i))
+    argmin = np.argmin(entropys)
+    bestweight = weight[indexs[argmin]]
+    return bestweight
 
 
-
-
-
-
-weight = np.load(CurrentPath+"WeightSearch.npy")
-# weight = np.zeros((990,3,3));
-# weight[:,1]=100
-# weight[:,:,1]=100
-utils.NumpyToImage(weight,CurrentPath+"image")
-
-a=0
-
-
-
-
-
-
-
-
-# weight = UnsuperLearnSearchWeight(model, layer, traindata, NumSearch=100000, SearchChannelRatio=8, Interation=10)
-# np.save("WeightSearch.npy", weight)
-
-weight = UnsuperLearnTrainWeight(model, layer, traindata)
-np.save("WeightTrain.npy", weight)
-
-
-
-ConvKernelToImage(weight, CurrentPath+"image")
-# utils.SaveModel(model,CurrentPath+"/checkpoint.pkl")
-print("save model sucess")
+def SetModelConvWeight(model, layer, weight):
+    w = utils.SetDevice(torch.from_numpy(weight))
+    model.features[layer].weight.data = w

FilterEvaluator/TrainNetwork.py

@@ -66,6 +66,8 @@ for i in range(1000):
     window.AppendData(linePretrain,Train.test(model,testdata))
 
 
+
+
 model = utils.SetDevice(Model.Net3335())
 model = utils.LoadModel(model, CurrentPath+"/checkpoint.pkl")
 

tools/utils.py

@@ -9,6 +9,7 @@ from easydict import EasyDict as edict
 import yaml
 import numpy as np
 import argparse
+import cv2
 
 class AverageMeter(object):
     """ Computes ans stores the average and current value"""
@@ -103,7 +104,7 @@ def SaveModel(model , filename='checkpoint_'+str(time.time()) + '.pkl'):
             # 'loss': loss,
             }, filename)
 def LoadModel(model, filename):
-    checkpoint = torch.load(filename)
+    checkpoint = torch.load(filename, map_location="cpu")
     model.load_state_dict(checkpoint['model_state_dict'])
     # optimizer.load_state_dict(checkpoint['optimizer_state_dict'])
     # epoch = checkpoint['epoch']
@@ -164,7 +165,8 @@ def ConvKernelToImage(model, layer, foldname):
         d = d.astype(int)
         cv2.imwrite(foldname+"/"+str(i)+".png", d)
 
-def NumpyToImage(numpydate, foldname ,maxImageWidth = 128,maxImageHeight = 128):
+
+def NumpyToImage(numpydate, foldname, title="", maxImageWidth=128, maxImageHeight=128):
     if not os.path.exists(foldname):
         os.mkdir(foldname)
     numpydatemin = np.min(numpydate)
@@ -187,5 +189,5 @@ def NumpyToImage(numpydate, foldname ,maxImageWidth = 128,maxImageHeight = 128):
         d=np.reshape(d, (imagerows, imagecols, datashape[1], datashape[2]))
         d=np.swapaxes(d, 1, 2)
         d=np.reshape(d, (imagerows*datashape[1],imagecols*datashape[2]))
-        d = d.astype(int)
-        cv2.imwrite(foldname+"/"+str(i)+"-"+str(i+stepimages)+".png", d)
+        d = d.astype("uint8")
+        cv2.imwrite(foldname+"/"+title+str(i)+"-"+str(i+stepimages)+".png", d)