tensorflow_glasses_classifier_plus_tflite

Transfer learning with Inception and MobileNet was used as out-of-box solution with options for rapid experimentation in limited time. After series of trials MobileNet was chosen for better inference time and requirement <3mb for model with low latency (compressed to tflite, realtime <50ms on Mali-T880 MP12). Although more data is required for better accuracy.

Launching

git clone https://github.com/ZackPashkin/tensorflow_glasses_classifier_plus_tflite
cd tensorflow_glasses_classifier_plus_tflite

# Put your data in tf_files/dataset
# MobileNet is available 0.25; 0,5; 0.75 and 1.0
# Image size can be 128,160,192, 224 pixels.

IMAGE_SIZE=224 
ARCHITECTURE="mobilenet_0.25_${IMAGE_SIZE}"
  
python -m scripts.retrain \
  --bottleneck_dir=tf_files/bottlenecks \
  --how_many_training_steps=5000 \
  --model_dir=tf_files/models/ \
  --summaries_dir=tf_files/training_summaries/"${ARCHITECTURE}" \
  --output_graph=tf_files/retrained_graph.pb \
  --output_labels=tf_files/retrained_labels.txt \
  --architecture="${ARCHITECTURE}" \
  --image_dir=tf_files/dataset 
  

#make prediction
#change image adding name from your dataset
 python -m scripts.label_image \
    --graph=tf_files/retrained_graph.pb  \
    --image=tf_files/dataset/without_glasses/000004.jpg \
    --labels=tf_files/retrained_labels.txt

#Compress model converting to TFlite format

#IMAGE_SIZE=224
!tflite_convert \
  --graph_def_file=tf_files/retrained_graph.pb \
  --output_file=tf_files/optimized_graph025_224.lite \
  --input_format=TENSORFLOW_GRAPHDEF \
  --output_format=TFLITE \
  --input_shape=1,224,224,3 \
  --input_array=input \
  --output_array=final_result \
  --inference_type=FLOAT \
  --input_data_type=FLOAT

Colab Notebook

Run in Google Colab

MobileNet 0,25 with 224 image size was used to make 1.9mb tflite model then imported on android Dataset: See with_glasses.zip without_glasses.zip Images from CelebA dataset were used Demo app:apk Possible improvements: Add more data (now it's only <150images/label) MobileNetV2 architecture