Multilabel image classification to classify movie poster and find appropriate movie genre with movie poster. For this project tensorflow framwork is used for deep learning model. The model is converted into tf lite to reduce final model size. This is a complete project with server side code and also fronted. To deploy deep learning model, flask is used and deployed into heroku server. By using Flutter framework frontend web app is developed and deployed into firebase.
Notebook: Click here
Web App: Movie Poster Classifier
Framework: Tensorflow 2.1
Datasets: Movie_Poster_Dataset
Input Shape: (350, 350, 3)
Label: (24)
Output Shape: (24)
Model Summary
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d (Conv2D) (None, 348, 348, 16) 448
_________________________________________________________________
batch_normalization (BatchNo (None, 348, 348, 16) 64
_________________________________________________________________
dropout (Dropout) (None, 348, 348, 16) 0
_________________________________________________________________
conv2d_1 (Conv2D) (None, 346, 346, 32) 4640
_________________________________________________________________
batch_normalization_1 (Batch (None, 346, 346, 32) 128
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 173, 173, 32) 0
_________________________________________________________________
dropout_1 (Dropout) (None, 173, 173, 32) 0
_________________________________________________________________
conv2d_2 (Conv2D) (None, 171, 171, 64) 18496
_________________________________________________________________
batch_normalization_2 (Batch (None, 171, 171, 64) 256
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 85, 85, 64) 0
_________________________________________________________________
dropout_2 (Dropout) (None, 85, 85, 64) 0
_________________________________________________________________
conv2d_3 (Conv2D) (None, 83, 83, 128) 73856
_________________________________________________________________
batch_normalization_3 (Batch (None, 83, 83, 128) 512
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 41, 41, 128) 0
_________________________________________________________________
dropout_3 (Dropout) (None, 41, 41, 128) 0
_________________________________________________________________
flatten (Flatten) (None, 215168) 0
_________________________________________________________________
dense (Dense) (None, 128) 27541632
_________________________________________________________________
batch_normalization_4 (Batch (None, 128) 512
_________________________________________________________________
dropout_4 (Dropout) (None, 128) 0
_________________________________________________________________
dense_1 (Dense) (None, 128) 16512
_________________________________________________________________
batch_normalization_5 (Batch (None, 128) 512
_________________________________________________________________
dropout_5 (Dropout) (None, 128) 0
_________________________________________________________________
dense_2 (Dense) (None, 24) 3096
=================================================================
Total params: 27,660,664
Trainable params: 27,659,672
Non-trainable params: 992
| Parameters | |
|---|---|
| Optimizer | Adam |
| Learning rate | 0.001 |
| Loss function | Binary Crossentropy |
| Batch size | 32 |
| Epoch | 5 |
Train on 6854 samples, validate on 1210 samples
Epoch 1/5
6854/6854 [==============================] - 39s 6ms/sample - loss: 0.6711 - accuracy: 0.6647 - val_loss: 0.3302 - val_accuracy: 0.9021
Epoch 2/5
6854/6854 [==============================] - 34s 5ms/sample - loss: 0.3010 - accuracy: 0.8932 - val_loss: 0.2634 - val_accuracy: 0.9020
Epoch 3/5
6854/6854 [==============================] - 34s 5ms/sample - loss: 0.2676 - accuracy: 0.9039 - val_loss: 0.2620 - val_accuracy: 0.9027
Epoch 4/5
6854/6854 [==============================] - 34s 5ms/sample - loss: 0.2602 - accuracy: 0.9062 - val_loss: 0.2636 - val_accuracy: 0.9030
Epoch 5/5
6854/6854 [==============================] - 34s 5ms/sample - loss: 0.2555 - accuracy: 0.9072 - val_loss: 0.2640 - val_accuracy: 0.9030
Tensor flow model is converted into tf.lite. The main model size is 350 mb+, after convert into tf lite the final size ~ 55 mb+.
converter = tf.lite.TFLiteConverter.from_keras_model(net)
tflite_model = converter.convert()This conversion reduce model size into ~110 mb.
Next, Float32 converted into Float 16.
converter.optimizations = [tf.lite.Optimize.DEFAULT]
converter.target_spec.supported_types = [tf.float16]
tflite_fp16_model = converter.convert()
And saved optimized model
tflite_model_fp16_file = tflite_models_dir/"movie_optimized.tflite"
tflite_model_fp16_file.write_bytes(tflite_fp16_model)
Now model size is 55.00 mb
Compare outputs between 3 models (tensorflow model, tflite model, tflite optimized model)
Output from tensorflow model:, War, Action, News
Output from tensorflow lite model:, War, Action, News
Output from tensorflow lite optimized model:, War, Action, News
Notebook link: Click here
For deploying model Flask is used
Used libraries
tensorflow
numpy
Flask==1.1.1
gunicorn==20.0.4
Pillow
flask-cors
Server accept post request with image bytes as data and return the predicted values in json format.
@app.route('/analysis/', methods=['POST', 'GET'])
def analysis():
if request.method == 'POST':
img = request.data
response = predict(img)
data = {
"result": response
}
return jsonify(
data)
Server link: https://rocky-beyond-88001.herokuapp.com/analysis
For frontend web app, Flutter web framework is used
Flutter 1.15.19
Dart 2.8.0
http: ^0.12.0+4
html: ^0.14.0+3
retrofit: ^1.3.1+1
logger: ^0.8.3
states_rebuilder: ^1.14.2
json_annotation: ^3.0.1
url_launcher: ^5.4.2
Website deployed in Firebase
Web app link: Movie Poster Classifier
Clone this repo
git clone https://github.com/Iamsdt/MultilabelMoviePosterClassification.git
For AI model see this notebook, click here
Server side code: all the server side code including model file located into server folder
Flutter Web App: Located into app/web_app
Open in the IDE with flutter plugin and run this command, to generate
flutter run -d chrome
Note: you need chorme web browser to run locally.
This project is under Apache License 2.0.
Copyright {2020} {Shudipto Trafder}
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.