search

nisargptl / vehicle-recognition

This enables users to gather information about any vehicle and they can see the details in a matter of seconds. Snap a quick photo in the app and it will provide the details about the vehicle make, model, year, user ratings and starting retail price. In addition, it will provide this same information for the 2 closest competitors so a user can do some quick comparison shopping.
https://viris.herokuapp.com/
0 stars 1 forks source link

Model2 #11

Closed khyatibhuva closed 3 years ago

khyatibhuva commented 3 years ago

Try adding few more classes (make it 100) and calculate efficiency. Data will be given by database team.

khyatibhuva commented 3 years ago

The database team will soon upload the updated spreadsheet. So use that data and try to improve accuracy.

sjdhola commented 3 years ago

Now we are starting to train model for 100 classes.

khyatibhuva commented 3 years ago

Noted. Update the team with accuracy also after you are done.

sjdhola commented 3 years ago

The execution of code is done. Here is the results we got for 100 cars.

Layer (type) Output Shape Param #

input_1 (InputLayer) [(None, 227, 227, 3)] 0

conv1 (Conv2D) (None, 114, 114, 32) 864

conv1_bn (BatchNormalization (None, 114, 114, 32) 128

conv1_relu (ReLU) (None, 114, 114, 32) 0

conv_dw_1 (DepthwiseConv2D) (None, 114, 114, 32) 288

conv_dw_1_bn (BatchNormaliza (None, 114, 114, 32) 128

conv_dw_1_relu (ReLU) (None, 114, 114, 32) 0

conv_pw_1 (Conv2D) (None, 114, 114, 64) 2048

conv_pw_1_bn (BatchNormaliza (None, 114, 114, 64) 256

conv_pw_1_relu (ReLU) (None, 114, 114, 64) 0

conv_pad_2 (ZeroPadding2D) (None, 115, 115, 64) 0

conv_dw_2 (DepthwiseConv2D) (None, 57, 57, 64) 576

conv_dw_2_bn (BatchNormaliza (None, 57, 57, 64) 256

conv_dw_2_relu (ReLU) (None, 57, 57, 64) 0

conv_pw_2 (Conv2D) (None, 57, 57, 128) 8192

conv_pw_2_bn (BatchNormaliza (None, 57, 57, 128) 512

conv_pw_2_relu (ReLU) (None, 57, 57, 128) 0

conv_dw_3 (DepthwiseConv2D) (None, 57, 57, 128) 1152

conv_dw_3_bn (BatchNormaliza (None, 57, 57, 128) 512

conv_dw_3_relu (ReLU) (None, 57, 57, 128) 0

conv_pw_3 (Conv2D) (None, 57, 57, 128) 16384

conv_pw_3_bn (BatchNormaliza (None, 57, 57, 128) 512

conv_pw_3_relu (ReLU) (None, 57, 57, 128) 0

conv_pad_4 (ZeroPadding2D) (None, 58, 58, 128) 0

conv_dw_4 (DepthwiseConv2D) (None, 28, 28, 128) 1152

conv_dw_4_bn (BatchNormaliza (None, 28, 28, 128) 512

conv_dw_4_relu (ReLU) (None, 28, 28, 128) 0

conv_pw_4 (Conv2D) (None, 28, 28, 256) 32768

conv_pw_4_bn (BatchNormaliza (None, 28, 28, 256) 1024

conv_pw_4_relu (ReLU) (None, 28, 28, 256) 0

conv_dw_5 (DepthwiseConv2D) (None, 28, 28, 256) 2304

conv_dw_5_bn (BatchNormaliza (None, 28, 28, 256) 1024

conv_dw_5_relu (ReLU) (None, 28, 28, 256) 0

conv_pw_5 (Conv2D) (None, 28, 28, 256) 65536

conv_pw_5_bn (BatchNormaliza (None, 28, 28, 256) 1024

conv_pw_5_relu (ReLU) (None, 28, 28, 256) 0

conv_pad_6 (ZeroPadding2D) (None, 29, 29, 256) 0

conv_dw_6 (DepthwiseConv2D) (None, 14, 14, 256) 2304

conv_dw_6_bn (BatchNormaliza (None, 14, 14, 256) 1024

conv_dw_6_relu (ReLU) (None, 14, 14, 256) 0

conv_pw_6 (Conv2D) (None, 14, 14, 512) 131072

conv_pw_6_bn (BatchNormaliza (None, 14, 14, 512) 2048

conv_pw_6_relu (ReLU) (None, 14, 14, 512) 0

conv_dw_7 (DepthwiseConv2D) (None, 14, 14, 512) 4608

conv_dw_7_bn (BatchNormaliza (None, 14, 14, 512) 2048

conv_dw_7_relu (ReLU) (None, 14, 14, 512) 0

conv_pw_7 (Conv2D) (None, 14, 14, 512) 262144

conv_pw_7_bn (BatchNormaliza (None, 14, 14, 512) 2048

conv_pw_7_relu (ReLU) (None, 14, 14, 512) 0

conv_dw_8 (DepthwiseConv2D) (None, 14, 14, 512) 4608

conv_dw_8_bn (BatchNormaliza (None, 14, 14, 512) 2048

conv_dw_8_relu (ReLU) (None, 14, 14, 512) 0

conv_pw_8 (Conv2D) (None, 14, 14, 512) 262144

conv_pw_8_bn (BatchNormaliza (None, 14, 14, 512) 2048

conv_pw_8_relu (ReLU) (None, 14, 14, 512) 0

conv_dw_9 (DepthwiseConv2D) (None, 14, 14, 512) 4608

conv_dw_9_bn (BatchNormaliza (None, 14, 14, 512) 2048

conv_dw_9_relu (ReLU) (None, 14, 14, 512) 0

conv_pw_9 (Conv2D) (None, 14, 14, 512) 262144

conv_pw_9_bn (BatchNormaliza (None, 14, 14, 512) 2048

conv_pw_9_relu (ReLU) (None, 14, 14, 512) 0

conv_dw_10 (DepthwiseConv2D) (None, 14, 14, 512) 4608

conv_dw_10_bn (BatchNormaliz (None, 14, 14, 512) 2048

conv_dw_10_relu (ReLU) (None, 14, 14, 512) 0

conv_pw_10 (Conv2D) (None, 14, 14, 512) 262144

conv_pw_10_bn (BatchNormaliz (None, 14, 14, 512) 2048

conv_pw_10_relu (ReLU) (None, 14, 14, 512) 0

conv_dw_11 (DepthwiseConv2D) (None, 14, 14, 512) 4608

conv_dw_11_bn (BatchNormaliz (None, 14, 14, 512) 2048

conv_dw_11_relu (ReLU) (None, 14, 14, 512) 0

conv_pw_11 (Conv2D) (None, 14, 14, 512) 262144

conv_pw_11_bn (BatchNormaliz (None, 14, 14, 512) 2048

conv_pw_11_relu (ReLU) (None, 14, 14, 512) 0

conv_pad_12 (ZeroPadding2D) (None, 15, 15, 512) 0

conv_dw_12 (DepthwiseConv2D) (None, 7, 7, 512) 4608

conv_dw_12_bn (BatchNormaliz (None, 7, 7, 512) 2048

conv_dw_12_relu (ReLU) (None, 7, 7, 512) 0

conv_pw_12 (Conv2D) (None, 7, 7, 1024) 524288

conv_pw_12_bn (BatchNormaliz (None, 7, 7, 1024) 4096

conv_pw_12_relu (ReLU) (None, 7, 7, 1024) 0

conv_dw_13 (DepthwiseConv2D) (None, 7, 7, 1024) 9216

flatten (Flatten) (None, 50176) 0

dense (Dense) (None, 100) 5017700

Total params: 7,189,796 Trainable params: 6,090,852 Non-trainable params: 1,098,944

Found 18885 images belonging to 100 classes. Found 2000 images belonging to 100 classes. Epoch 1/25 74/74 [==============================] - 10147s 137s/step - loss: 61.9345 - accuracy: 0.0264 - val_loss: 65.1034 - val_accuracy: 0.0400 Epoch 2/25 74/74 [==============================] - 216s 3s/step - loss: 3.8809 - accuracy: 0.3338 - val_loss: 7.7157 - val_accuracy: 0.2645 Epoch 3/25 74/74 [==============================] - 215s 3s/step - loss: 0.6054 - accuracy: 0.8277 - val_loss: 1.7625 - val_accuracy: 0.5865 Epoch 4/25 74/74 [==============================] - 215s 3s/step - loss: 0.2127 - accuracy: 0.9387 - val_loss: 1.2184 - val_accuracy: 0.7365 Epoch 5/25 74/74 [==============================] - 215s 3s/step - loss: 0.1133 - accuracy: 0.9647 - val_loss: 0.9882 - val_accuracy: 0.7915 Epoch 6/25 74/74 [==============================] - 215s 3s/step - loss: 0.0945 - accuracy: 0.9730 - val_loss: 1.0764 - val_accuracy: 0.7995 Epoch 7/25 74/74 [==============================] - 215s 3s/step - loss: 0.0838 - accuracy: 0.9790 - val_loss: 1.1060 - val_accuracy: 0.8075 Epoch 8/25 74/74 [==============================] - 215s 3s/step - loss: 0.0486 - accuracy: 0.9865 - val_loss: 0.7437 - val_accuracy: 0.8680 Epoch 9/25 74/74 [==============================] - 215s 3s/step - loss: 0.0159 - accuracy: 0.9959 - val_loss: 0.7195 - val_accuracy: 0.8730 Epoch 10/25 74/74 [==============================] - 215s 3s/step - loss: 0.0121 - accuracy: 0.9969 - val_loss: 0.7039 - val_accuracy: 0.8770 Epoch 11/25 74/74 [==============================] - 215s 3s/step - loss: 0.0086 - accuracy: 0.9984 - val_loss: 0.7183 - val_accuracy: 0.8745 Epoch 12/25 74/74 [==============================] - 215s 3s/step - loss: 0.0119 - accuracy: 0.9973 - val_loss: 0.7149 - val_accuracy: 0.8795 Epoch 13/25 74/74 [==============================] - 215s 3s/step - loss: 0.0065 - accuracy: 0.9986 - val_loss: 0.7178 - val_accuracy: 0.8795 Epoch 14/25 74/74 [==============================] - 215s 3s/step - loss: 0.0096 - accuracy: 0.9982 - val_loss: 0.7181 - val_accuracy: 0.8790 Epoch 15/25 74/74 [==============================] - 215s 3s/step - loss: 0.0087 - accuracy: 0.9980 - val_loss: 0.7192 - val_accuracy: 0.8790 Epoch 16/25 74/74 [==============================] - 215s 3s/step - loss: 0.0078 - accuracy: 0.9985 - val_loss: 0.7196 - val_accuracy: 0.8790 Epoch 17/25 74/74 [==============================] - 214s 3s/step - loss: 0.0083 - accuracy: 0.9982 - val_loss: 0.7196 - val_accuracy: 0.8790

8/8 [==============================] - 6s 698ms/step - loss: 0.7196 - accuracy: 0.8790 [0.7196163535118103, 0.8790000081062317]

sjdhola commented 3 years ago

From the simulation results we are getting 87.90% accuracy for 100 car model

khyatibhuva commented 3 years ago

Noted.