Deep-Model-Transfer-Deployment

After comparing the accuracy and speed of several models, I finally select Inception-v3 to deploy my 102 classes flower classification project which was trained on Oxford-102 flower dataset. The final model file in this folder which include the freezed pb file.

Other project including 🕊️Bird-200, 🚗Car-196, 🐶Dog-120, 🐶🐱Pet-37 described in Deep-Model-Transfer will be enclosed in the future.

Environment

GTX 1080
CUDA 10.1
CuDNN 7.6.5
TensorFlow 1.5.0+/2.0.0+

Prepare

For TensorFlow 2.0.0+

nothing to change

For TensorFlow 1.5.0+

comment import tensorflow.compat.v1 as tf

uncomment import tensorflow as tf

same as server_html.py/server_api.py

Through HTML

Command

# create folder to store images uploaded
>>> mkdir /tem/upload
>>> python3 -u server_html.py \
    --model_name=./all_102_inception_v3/all_102_inception_v3_named_freeze.pb \
    --label_file=./all_102_inception_v3/all_102_inception_v3_named_freeze.label \
    --upload_folder=/tmp/upload

How open http://<your_ip>:5001/ upload image get result

Through API

Command

# create folder to store images uploaded
>>> mkdir /tem/upload
>>> python3 -u server_api.py \
    --model_name=./all_102_inception_v3/all_102_inception_v3_named_freeze.pb \
    --label_file=./all_102_inception_v3/all_102_inception_v3_named_freeze.label \
    --upload_folder=/tmp/upload

Definition

If upload image URL path:

POST /FROMURL

If upload image file:

POST /FROMFILE

Arguments