donniet / ptag_detector

ptag-detector

Version

TF Version : 2.5.0
Cuda : 11.4.1
Cudnn: 8.2.2.26
Nvidia: 471.68

Installation

Step 1: Partitioning the annotated images to test and train directory

Script-location: TensorFlow-2.0/scripts/pre-processing

python partition_dataset.py -x -i ../../workspace/training/images -r 0.2

Step 2: Generating TFRecord Test and Train

Script-location: TensorFlow-2.0/scripts/pre-processing

python generate_tfrecord.py -x D:\Workspace\TensorFlow-2.0\workspace\training\images\train -l D:\Workspace\TensorFlow-2.0\workspace\training\annotations\label_map.pbtxt -o D:\Workspace\TensorFlow-2.0\workspace\training\annotations\train.record
python generate_tfrecord.py -x D:\Workspace\TensorFlow-2.0\workspace\training\images\test -l D:\Workspace\TensorFlow-2.0\workspace\training\annotations\label_map.pbtxt -o D:\Workspace\TensorFlow-2.0\workspace\training\annotations\test.record

Step 3: Initiate Training

Script-location: TensorFlow-2.0/workspace/training

python model_main_tf2.py --model_dir=models/ssd_mobilenet_v2_fpnlite_320x320 --pipeline_config_path=models/ssd_mobilenet_v2_fpnlite_320x320/pipeline.config --checkpoint_dir=models/ssd_mobilenet_v2_fpnlite_320x320

Step 4: Export Model (Generate Saved_Model)

Script-location: TensorFlow-2.0/workspace/training

python .\exporter_main_v2.py --input_type image_tensor --pipeline_config_path D:\Workspace\TensorFlow-2.0\workspace\training\models\ssd_mobilenet_v2_fpnlite_320x320\pipeline.config --trained_checkpoint_dir D:\Workspace\TensorFlow-2.0\workspace\training\models\ssd_mobilenet_v2_fpnlite_320x320 --output_directory D:\Workspace\TensorFlow-2.0\workspace\training\exported-models\ptag-detector-model

Step 5: Convert Saved Model to TFLite

Script-location: TensorFlow-2.0/workspace/training

python export_tflite_graph_tf2.py --pipeline_config_path D:\Workspace\TensorFlow-2.0\workspace\training\models\ssd_mobilenet_v2_fpnlite_320x320\pipeline.config --trained_checkpoint_dir D:\Workspace\TensorFlow-2.0\workspace\training\models\ssd_mobilenet_v2_fpnlite_320x320 --output_directory D:\Workspace\TensorFlow-2.0\workspace\training\exported-models\ptag-detector-model

Step 6: Post Training Quantization(This Script may not work properly for TF 2.5.0, Please refer to Tensorflow website)

Script-location: TensorFlow-2.0/workspace/training

tflite_convert --graph_def_file=tflite_model/tflite_graph.pb --output_file=tflite_model/detect.tflite --output_format=TFLITE --input_shapes=1,300,300,3 --input_arrays=normalized_input_image_tensor --output_arrays='TFLite_Detection_PostProcess','TFLite_Detection_PostProcess:1','TFLite_Detection_PostProcess:2','TFLite_Detection_PostProcess:3' --inference_type=QUANTIZED_UINT8 --mean_values=128 --std_dev_values=127 --change_concat_input_ranges=false --allow_custom_ops

Step 7: Convert to edgetpu compatible

docker run -ti -v e:/:/data ubuntu /bin/bash
:/usr/bin# edgetpu_compiler ../../tflite_model/detect.tflite -o ../../