Update resnet50_imagenet.py

llvm/projects/keras/src/resnet50_imagenet.py

@@ -11,12 +11,12 @@ import tensorflow as tf
 import keras
 from keras.models import Sequential, Model
 from keras.layers import *
-from keras.applications.resnet50 import ResNet50, preprocess_input
 from keras.utils import to_categorical
+from keras.applications.resnet50 import ResNet50, preprocess_input
 from keras import backend as K
 
 from frontend.approxhpvm_translator import translate_to_approxhpvm
-from frontend.weight_utils import dumpCalibrationData
+from frontend.weight_utils import dumpCalibrationData2
 
 
 np.random.seed(2020)
@@ -29,29 +29,13 @@ data_format = 'channels_first'
 
 
 IMAGENET_DIR = '/home/nz11/ILSVRC2012/'
-OUTPUT_DIR = 'data/resnet50_imagenet/'
-
-NUM_CLASSES = 100
-IMAGES_PER_CLASS = 200
-VAL_SIZE = 100
+OUTPUT_DIR = 'data/resnet50_imagenet_tune/'
+WEIGHTS_PATH = 'data/resnet50_imagenet/weights.h5'
 
+NUM_CLASSES = 200
+IMAGES_PER_CLASS = 40
+# VAL_SIZE = 100
 
-# def get_resnet50_nchw_keras():
-    
-#     model = ResNet50()
-    
-#     for x in model.layers:
-#         print (x.name)
-    
-#     x = model.get_layer('flatten_1').output
-#     x = Dense(1000, name='fc1000')(x)
-#     x = Activation('softmax')(x)
-    
-#     model_nchw = Model(model.input, x)
-    
-#     model_nchw.get_layer('fc1000').set_weights(model.get_layer('fc1000').get_weights())
-    
-#     return model_nchw
 
 
 def identity_block(input_tensor, kernel_size, filters, stage, block):
@@ -205,8 +189,8 @@ f.close()
 
 model = get_resnet50_nchw_keras()
     
-X_test = []
-y_true = []
+X_tune, X_test = [], []
+y_tune, y_true = [], []
     
 classes = glob.glob(IMAGENET_DIR + 'val/*')
 
@@ -217,13 +201,21 @@ for c in np.random.permutation(len(classes))[:NUM_CLASSES]:
     idx = np.random.permutation(len(x))
     idx = idx[:max(len(idx), IMAGES_PER_CLASS)]
     
-    X_test += list(map(lambda x : load_image(x), x[idx]))
-    
     synset = classes[c].split('/')[-1]
-    y_true += [synset_to_keras_idx[synset]] * len(x[idx])
+    images = list(map(lambda x : load_image(x), x[idx]))
+    labels = [synset_to_keras_idx[synset]] * len(x[idx])
+    
+    X_test += images[:IMAGES_PER_CLASS // 2]
+    y_true += labels[:IMAGES_PER_CLASS // 2]
+    
+    X_tune += images[IMAGES_PER_CLASS // 2:]
+    y_tune += labels[IMAGES_PER_CLASS // 2:]
+    
     
 X_test = np.array(X_test)
 y_true = np.array(y_true)
+X_tune = np.array(X_tune)
+y_tune = np.array(y_tune)
 
 
 
@@ -247,7 +239,6 @@ def train_helper(x):
 
 
 
-
 train_images = glob.glob(IMAGENET_DIR + 'train/*/*')
 random.shuffle(train_images)
 
@@ -258,7 +249,7 @@ dataset = dataset.map(
 )
 
 dataset = dataset.shuffle(buffer_size=1000)
-dataset = dataset.batch(32)
+dataset = dataset.batch(64)
 dataset = dataset.repeat()
 
 next_element = dataset.make_one_shot_iterator().get_next()
@@ -273,15 +264,22 @@ def generate():
 
 model.compile(optimizer=keras.optimizers.Adam(lr=0.00001), loss='categorical_crossentropy', metrics=['acc'])
 
+if os.path.exists(WEIGHTS_PATH):
+    model.load_weights(WEIGHTS_PATH)
+else:
+    model.fit_generator(generate(), steps_per_epoch=1000, validation_data=(X_test, to_categorical(y_true, num_classes=1000)), epochs=6)
+    model.save_weights(OUTPUT_DIR + 'weights.h5')
 
-model.fit_generator(generate(), steps_per_epoch=1000, validation_data=(X_test, to_categorical(y_true, num_classes=1000)), epochs=6)
-
+translate_to_approxhpvm(model, OUTPUT_DIR, X_tune, y_tune, 1000)
 
-translate_to_approxhpvm(model, OUTPUT_DIR, X_test[:VAL_SIZE], y_true[:VAL_SIZE], 1000)
+# dumpCalibrationData2(OUTPUT_DIR + 'test_input_10K.bin', X_test, OUTPUT_DIR + 'test_labels_10K.bin', y_true)
+dumpCalibrationData2(OUTPUT_DIR + 'tune_input.bin', X_tune, OUTPUT_DIR + 'tune_labels.bin', y_tune)
+dumpCalibrationData2(OUTPUT_DIR + 'test_input.bin', X_test, OUTPUT_DIR + 'test_labels.bin', y_true)
 
-dumpCalibrationData(OUTPUT_DIR + 'test_input.bin', X_test, OUTPUT_DIR + 'test_labels.bin', y_true)
 
-
-# pred = np.argmax(model.predict(X_test), axis=1)
-# print ('val accuracy', np.sum(pred == y_true.ravel()) / len(X_test))
+pred = np.argmax(model.predict(X_test), axis=1)
+print ('val accuracy', np.sum(pred == y_true.ravel()) / len(X_test))
+    
+pred = np.argmax(model.predict(X_tune), axis=1)
+print ('val accuracy', np.sum(pred == y_tune.ravel()) / len(X_tune))
     
\ No newline at end of file