Refine kernal Random to w*h*16 kinds possiable.

Donot use Cpu any more.

FilterEvaluator/Evaluator.py

@@ -16,7 +16,6 @@ import random
 import cv2
 
 
-
 CurrentPath = os.path.split(os.path.realpath(__file__))[0]+"/"
 print("Current Path :" + CurrentPath)
 

FilterEvaluator/EvaluatorUnsuper.py

@@ -16,6 +16,7 @@ import random
 import cv2
 from tqdm import tqdm
 import threading
+from itertools import combinations
 
 
 CurrentPath = os.path.split(os.path.realpath(__file__))[0]+"/"
@@ -26,6 +27,19 @@ import Model as Model
 from tools import utils, Train, Loader
 
 
+
+
+
+
+# iterable = combinations(range(72), 4)
+# count = iterable.count()
+# pool = tuple(iterable)
+# n = len(pool)
+# indices = sorted(random.sample(range(n), 2))
+# fdafda = tuple(pool[i] for i in indices)
+
+
+
 def GetScore(netmodel,layer,SearchLayer,DataSet):
     netmodel.eval()
     sample = utils.SetDevice(torch.empty((SearchLayer.out_channels,0)))
@@ -184,10 +198,12 @@ def UnsuperLearnFindBestWeight(netmodel, layer, weight, dataloader, databatchs=1
 
 
 # search best weight from random data
-def UnsuperLearnSearchBestWeight(netmodel, layer, dataloader, databatchs=128, stepsize=1000, interation=1000):
+# Test total sample size = stepsize * stepsize * interation
+# Random kernel number = in_channels * kernel_size[0] * kernel_size[1] * 16
+def UnsuperLearnSearchBestWeight(netmodel, layer, dataloader, databatchs=128, stepsize=20, interation=1000):
     interationbar = tqdm(total=interation)    
     forwardlayer = layer -1
-
+    samplesize = stepsize*stepsize
     netmodel.eval()
     tl = netmodel.features[layer]
     outchannels = tl.out_channels
@@ -211,93 +227,55 @@ def UnsuperLearnSearchBestWeight(netmodel, layer, dataloader, databatchs=128, st
     for i in range(outchannels):
         shift.append(1<<i)
     shift = utils.SetDevice(torch.from_numpy(np.array(shift).astype("uint8")))
-    # shift = torch.from_numpy(np.array(shift).astype("uint8"))
-    
+
+    newweightshape = list(newlayer.weight.data.shape)
+    samplekernelsize = newlayer.in_channels*newlayer.kernel_size[0]*newlayer.kernel_size[1]*16
+    newweightshape[0] = samplekernelsize
+    newweight = np.random.uniform(-1.0,1.0,newweightshape).astype("float32")
+    newweight = newweight.reshape((-1,newweightshape[-1]*newweightshape[-2]))
+    newweight = np.swapaxes(newweight,0,1)-np.mean(newweight,-1)
+    newweight = np.swapaxes(newweight,0,1).reshape(newweightshape)
+    newlayer.weight.data=utils.SetDevice(torch.from_numpy(newweight))
+    outputs = newlayer(datasnetout).transpose(0,1)
+
+    indexs = np.random.randint(0, samplekernelsize, samplesize*interation*outchannels).reshape(interation,samplesize, -1)
+
+    # da = np.random.randint(0,5,2*9).reshape(9,2)
+    # te = np.sort(da,1)
+    # tg = np.unique(te ,axis=0)
+    # list(combinations(range(newweightshape[0]), 8))
+
     bestweight = []
     bestentropy = [100.0]
-    bittedset = []
-    bittedLock = threading.Lock()
-    bestLock = threading.Lock()
-
-    class CPU (threading.Thread):  
-        def __init__(self):
-            threading.Thread.__init__(self)
-        def run(self):
-            hasdata = 0
-            bittedLock.acquire()
-            if len(bittedset)>0:
-                bitted = bittedset.pop()
-                hasdata = 1
-            bittedLock.release()
-            if hasdata > 0:
-                entropys = []
-                for i in range(len(indexs)):
-                    histced = bitted[:,i].histc(256,0,255).type(torch.float32)
-                    histced = histced[histced>0]
-                    histced = histced/histced.sum()
-                    entropy = (histced.log2()*histced).sum()
-                    entropys.append(entropy.numpy())
-                argmin = np.argmin(entropys)
-
-                bestLock.acquire()
-                if entropys[argmin] < bestentropy[0]:
-                    bestweight = newweight[indexs[argmin]]
-                    bestentropy[0] = entropys[argmin]
-                    print("finded better entropy")
-                bestLock.release()
 
     for j in range(interation):
-        newweightshape = list(newlayer.weight.data.shape)
-        newweightshape[0] = stepsize
-        newweight = np.random.uniform(-1.0,1.0,newweightshape).astype("float32")
-        newweight = newweight.reshape((-1,newweightshape[-1]*newweightshape[-2]))
-        newweight = np.swapaxes(newweight,0,1)-np.mean(newweight,-1)
-        newweight = np.swapaxes(newweight,0,1).reshape(newweightshape)
-        
-        newlayer.weight.data=utils.SetDevice(torch.from_numpy(newweight))
-        outputs = newlayer(datasnetout).transpose(0,1)
-        samplesize = stepsize*stepsize
-        indexs = np.random.randint(0,stepsize,samplesize*outchannels).reshape(samplesize,-1)
-        # 1000 8 4096 5 5
-        outputs = outputs[indexs]
-        # 102400 1000 8        
-        reshaped = outputs.reshape(samplesize, outchannels, -1).permute(2, 0, 1)
-        # 1000 8
+        # 400 8 4096 5 5
+        output = outputs[indexs[j]]
+        # 102400 400 8        
+        reshaped = output.reshape(samplesize, outchannels, -1).permute(2, 0, 1)
+        # 400 8
         meaned = reshaped.mean(0)
-        # 102400 1000
-        bitted = ((reshaped > meaned)* shift).sum(2).type(torch.float32).detach().cpu()
+        # 102400 400
+        # bitted = ((reshaped > meaned)* shift).sum(2).type(torch.float32).detach().cpu()
+        bitted = ((reshaped > meaned)* shift).sum(2).type(torch.float32)
         
-        bittedLock.acquire()
-        bittedset.append(bitted)
-        bittedLock.release()
-        threadcpu = CPU()
-        threadcpu.start()
-
-        # entropys = []
-        # for i in range(len(indexs)):
-        #     histced = bitted[:,i].histc(256,0,255).type(torch.float32)
-        #     histced = histced[histced>0]
-        #     histced = histced/histced.sum()
-        #     entropy = (histced.log2()*histced).sum()
-        #     entropys.append(entropy.detach().cpu().numpy())
-        
-        # argmin = np.argmin(entropys)
-        # if entropys[argmin] < bestentropy:
-        #     bestweight = newweight[indexs[argmin]]
-        #     bestentropy = entropys[argmin]
-
+        entropys = []
+        for i in range(samplesize):
+            histced = bitted[:,i].histc(256,0,255).type(torch.float32)
+            histced = histced[histced>0]
+            histced = histced/histced.sum()
+            entropy = (histced.log2()*histced).sum()
+            entropys.append(entropy.detach().cpu().numpy())
+        argmin = np.argmin(entropys)
+        if entropys[argmin] < bestentropy:
+            bestweight = newweight[indexs[j][argmin]]
+            bestentropy = entropys[argmin]
         interationbar.update(1)
-        interationbar.set_description("left:"+str(len(bittedset)))
-
-    while bittedset:
-        time.sleep(100)
-    
+        interationbar.set_description("entropy:"+str(bestentropy))
     interationbar.close()
     return bestweight
 
 
-
-
 def SetModelConvWeight(model, layer, weight):
     w = utils.SetDevice(torch.from_numpy(weight))
     model.features[layer].weight.data = w