Adding VGG ImageNet

llvm/projects/keras/src/vgg16_imagenet.py

+import os
+import glob
+
+import scipy
+import scipy.io
+import cv2
+
+import keras
+from keras.models import Model
+from keras.layers import *
+from keras.applications.vgg16 import VGG16, preprocess_input
+from keras.utils import to_categorical
+
+from frontend.approxhpvm_translator import translate_to_approxhpvm
+from frontend.weight_utils import dumpCalibrationData
+
+data_format = 'channels_first'
+
+def get_vgg16_nchw_keras():
+    img_input = Input(shape=(3, 224, 224))
+    
+    # Block 1
+    x = ZeroPadding2D(padding=(1, 1), data_format=data_format)(img_input)
+    x = Conv2D(64, (3, 3),
+                      padding='valid',
+                      data_format=data_format)(x)
+    x = Activation('relu')(x)
+    x = ZeroPadding2D(padding=(1, 1), data_format=data_format)(x)
+    x = Conv2D(64, (3, 3),
+                      padding='valid',
+                      data_format=data_format)(x)
+    x = Activation('relu')(x)
+    x = MaxPooling2D((2, 2), strides=(2, 2), data_format=data_format)(x)
+
+    # Block 2
+    x = ZeroPadding2D(padding=(1, 1), data_format=data_format)(x)
+    x = Conv2D(128, (3, 3),
+                      padding='valid',
+                      data_format=data_format)(x)
+    x = Activation('relu')(x)
+    x = ZeroPadding2D(padding=(1, 1), data_format=data_format)(x)
+    x = Conv2D(128, (3, 3),
+                      padding='valid',
+                      data_format=data_format)(x)
+    x = Activation('relu')(x)
+    x = MaxPooling2D((2, 2), strides=(2, 2), data_format=data_format)(x)
+
+    # Block 3
+    x = ZeroPadding2D(padding=(1, 1), data_format=data_format)(x)
+    x = Conv2D(256, (3, 3),
+                      padding='valid',
+                      data_format=data_format)(x)
+    x = Activation('relu')(x)
+    x = ZeroPadding2D(padding=(1, 1), data_format=data_format)(x)
+    x = Conv2D(256, (3, 3),
+                      padding='valid',
+                      data_format=data_format)(x)
+    x = Activation('relu')(x)
+    x = ZeroPadding2D(padding=(1, 1), data_format=data_format)(x)
+    x = Conv2D(256, (3, 3),
+                      padding='valid',
+                      data_format=data_format)(x)
+    x = Activation('relu')(x)
+    x = MaxPooling2D((2, 2), strides=(2, 2), data_format=data_format)(x)
+
+    # Block 4
+    x = ZeroPadding2D(padding=(1, 1), data_format=data_format)(x)
+    x = Conv2D(512, (3, 3),
+                      padding='valid',
+                      data_format=data_format)(x)
+    x = Activation('relu')(x)
+    x = ZeroPadding2D(padding=(1, 1), data_format=data_format)(x)
+    x = Conv2D(512, (3, 3),
+                      padding='valid',
+                      data_format=data_format)(x)
+    x = Activation('relu')(x)
+    x = ZeroPadding2D(padding=(1, 1), data_format=data_format)(x)
+    x = Conv2D(512, (3, 3),
+                      padding='valid',
+                      data_format=data_format)(x)
+    x = Activation('relu')(x)
+    x = MaxPooling2D((2, 2), strides=(2, 2), data_format=data_format)(x)
+
+    # Block 5
+    x = ZeroPadding2D(padding=(1, 1), data_format=data_format)(x)
+    x = Conv2D(512, (3, 3),
+                      padding='valid',
+                      data_format=data_format)(x)
+    x = Activation('relu')(x)
+    x = ZeroPadding2D(padding=(1, 1), data_format=data_format)(x)
+    x = Conv2D(512, (3, 3),
+                      padding='valid',
+                      data_format=data_format)(x)
+    x = Activation('relu')(x)
+    x = ZeroPadding2D(padding=(1, 1), data_format=data_format)(x)
+    x = Conv2D(512, (3, 3),
+                      padding='valid',
+                      data_format=data_format)(x)
+    x = Activation('relu')(x)
+    x = MaxPooling2D((2, 2), strides=(2, 2), data_format=data_format)(x)
+
+    x = Flatten(data_format=data_format)(x)
+    
+    x = Dense(4096)(x)
+    x = Activation('relu')(x)
+    x = Dense(4096)(x)   
+    x = Activation('relu')(x)
+    x = Dense(1000)(x)
+    x = Activation('softmax')(x)
+
+    model = Model(img_input, x)
+    
+    return model
+
+
+def load_image(x):
+    try:
+        x = x.decode('utf-8')
+    except:
+        pass
+    
+    image = cv2.imread(x)
+        
+    height, width, _ = image.shape
+    new_height = height * 256 // min(image.shape[:2])
+    new_width = width * 256 // min(image.shape[:2])
+    image = cv2.resize(image, (new_width, new_height), interpolation=cv2.INTER_CUBIC)
+    
+    height, width, _ = image.shape
+    startx = width // 2 - (224 // 2)
+    starty = height // 2 - (224 // 2)
+    image = image[starty:starty + 224, startx:startx + 224]
+    
+    image = image[:, :, ::-1]
+    image = preprocess_input(image.astype(np.float32))
+    image = np.transpose(image, (2, 0, 1))
+        
+    return image.astype(np.float32)
+
+
+if __name__ == '__main__':
+
+    os.environ["CUDA_VISIBLE_DEVICES"] = "1"
+    
+    num_images = 5000
+    val_size = 100
+    
+    model = VGG16()
+    model_nchw = get_vgg16_nchw_keras()   
+    
+    j = 0
+    for i in range(len(model_nchw.layers)):
+        if 'padding' in model_nchw.layers[i].name or 'activation' in model_nchw.layers[i].name:
+            continue
+        try:
+            model_nchw.layers[i].set_weights(model.layers[j].get_weights())
+        except:
+            print (i, model_nchw.layers[i], 'skipped')
+        j += 1
+    
+    
+    classes = os.listdir('/home/nz11/ILSVRC2012/train')
+    train_images = glob.glob('/home/nz11/ILSVRC2012/train/*/*')
+    val_images = glob.glob('/home/nz11/ILSVRC2012/val/*/*')
+    val_images = sorted(val_images, key=lambda x: x.split('/')[-1].split('_')[-1].split('.')[0])
+
+    idx = np.random.permutation(len(val_images))[:num_images]
+    val_images = np.array(val_images)[idx]
+
+    d = {k:v for v, k in enumerate(classes)}
+    
+    
+
+    X_test = []
+    for x in val_images:
+        X_test.append(load_image(x))
+    X_test = np.array(X_test)    
+        
+    
+    meta = scipy.io.loadmat("/home/nz11/ILSVRC2012/ILSVRC2012_devkit_t12/data/meta.mat")
+    original_idx_to_synset = {}
+    synset_to_name = {}
+
+    for i in range(1000):
+        ilsvrc2012_id = int(meta["synsets"][i,0][0][0][0])
+        synset = meta["synsets"][i,0][1][0]
+        name = meta["synsets"][i,0][2][0]
+        original_idx_to_synset[ilsvrc2012_id] = synset
+        synset_to_name[synset] = name
+
+    synset_to_keras_idx = {}
+    keras_idx_to_name = {}
+    f = open("/home/nz11/ILSVRC2012/ILSVRC2012_devkit_t12/data/synset_words.txt","r")
+    c = 0
+    for line in f:
+        parts = line.split(" ")
+        synset_to_keras_idx[parts[0]] = c
+        keras_idx_to_name[c] = " ".join(parts[1:])
+        c += 1
+    f.close()
+
+    def convert_original_idx_to_keras_idx(idx):
+        return synset_to_keras_idx[original_idx_to_synset[idx]]
+    
+    
+    f = open("/home/nz11/ILSVRC2012/ILSVRC2012_devkit_t12/data/ILSVRC2012_validation_ground_truth.txt","r")
+    y_true = f.read().strip().split("\n")
+    y_true = list(map(int, y_true))
+    y_true = np.array([convert_original_idx_to_keras_idx(idx) for idx in y_true])[idx]
+    y_true = to_categorical(y_true, num_classes=1000)
+    f.close()
+    
+    
+    translate_to_approxhpvm(model_nchw, "data/vgg16_imagenet/", X_test[:val_size], y_true[:val_size], 1000)
+    
+    dumpCalibrationData("data/vgg16_imagenet/test_input.bin", X_test, "data/vgg16_imagenet/test_labels.bin", y_true)
+
+
+    
+    y_pred = model_nchw.predict(X_test[:val_size])
+    print ('val accuracy', np.sum(np.argmax(y_pred, axis=1) == np.argmax(y_true[:val_size], axis=1)) / val_size)
+    
+    
+    
+    
\ No newline at end of file