conver tensorflow graph to tflite graph and run it on both pc and raspberry pi — 3
Project detail
Hi All
I need help to convert the TensorFlow graph that I trained with the ssd_mobilenet_v1 model to tflite graph
i used tensorflow 1.14 for training
i used the following tutorial for training
https://tensorflow-object-detection-api-tutorial.readthedocs.io/en/tensorflow-1.14/training.html
the training passed successfully and working well with TensorFlow but when converting it to tflite stop working
i got the following files from training -> model.ckpt-81701*
i used the following commands for convert it to tflite
python ./../../models/research/object_detection/export_tflite_ssd_graph.py –pipeline_config_path=./training/pipeline.config –trained_checkpoint_prefix=./training/model.ckpt-81701 –output_directory=./tflite_model –add_postprocessing_op=true
after this command i got 2 files -> tflite_graph.pbtxt&tflite_graph.pb
then i run the following command
tflite_convert –output_file=./tflite_model/model.tflite –graph_def_file=./tflite_model/tflite_graph.pb –input_shapes=1,360,640,3 –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=FLOAT –mean_values=128 –std_dev_values=128 –change_concat_input_ranges=false –allow_custom_ops
after this command i got 1 file -> model.tflite
but when trying to run it I got just a lot of false detections
so I need you to convert the graph to tflite and run it on pc and raspberry pi
after you see that all working well you should send me what to do step by step to convert and run it on both pc with ubuntu OS and raspberry pi
also, I attached some videos for test after implementation
videos -> https://drive.google.com/drive/folders/1tv1vqI5CpUaxBXhZL32Ms_PpJABD6Vpq?usp=sharing
training files -> https://drive.google.com/drive/folders/1l5HRgKed0BeXuKC14MfvrprlPS41vZWR?usp=sharing
***the project should be completed in 48 hours, after that i will close the project without paying and move it to another freelancer***