This website uses a object detection model trained on tensorflow to identify one of five pokemon in images.
Run git clone and npm install to install the app. Run npm run to start the webpage on localhost.
for training a new model:
follow instructions here to install tenosrflow using verison 1.12 of tensorflow instead of 1.9.
follow these inside of the Tensorflow directory to train a custom model.
once the model is trained run
tensorflowjs_converter --input_format=tf_saved_model
--output_node_names='Postprocessor/ExpandDims_1,Postprocessor/Slice'
--output_json=true
--saved_model_tags=serve
./saved_model
./web_model
where saved model is the train model .pb file, and web_model is the output directory for the .json file and associated weights
load the model by uploading to github and copying the address to the raw file on git hub and pasting it in the model url.
paste an image url into the text if the image is displays, hit run and the app will process the image and display bounding box and label for the positive detections
hosted on surge here
- tensorflow - Machine Learning and Neural Network Library in Python used for training
- tensorflowjs - ML and NN library for javascript
- surge - Deployment platform
- node.js - package manager
- anacondapython machine learning package manager and virtual environment manager
- Kyle Clabough - Initial work - SirAirdude
See also the list of contributors who participated in this project.
Pokemon is property of Nintendo, Gamefreak, and The Pokemon Company International