search

ArunMichaelDsouza / tensorflow-image-detection

A generic image detection program that uses Google's Machine Learning library, Tensorflow and a pre-trained Deep Learning Convolutional Neural Network model called Inception.
MIT License
328 stars 88 forks source link

Adding support for SavedModel for converting model to tfjs_graph_model #12

Open binary-person opened 5 years ago

binary-person commented 5 years ago

The title says it all. Do you have time to make this enhancement?

ArunMichaelDsouza commented 5 years ago

Hi @scheng123, can you please explain what this enhancement would do? Also, I would really appreciate a PR if you have the time. 🙂

binary-person commented 5 years ago

Since I’m experienced in web development, I was planning to have the program export it’s own JavaScript model, so I can later on incorporate that into a browser, which would be cool! And also, since I’m new to python, I have no idea how to do that (tried everything), although it would be awesome if I could make a PR. Do you have the time? I always love learning new things. 😊

binary-person commented 5 years ago

Looks like its impossible since the inception model was using a deprecated BatchNormWithGlobalNormalization, and an unsupported DecodeJpeg. Nice try though. Anyway, apart from the fact that converting it to a javascript model won't work anyway, you want me to make a pull request for adding support for the checkpoint and SavedModel saving feature?

ArunMichaelDsouza commented 5 years ago

Yes ofcourse, that would be great!

binary-person commented 5 years ago

Added checkpoint, and SavedModel saving feature. Also added a CTRL+C save as an extra bonus. As my use of language differs from yours, can you update the README.md to update the ability to save with the combination of CTRL+C? As always, I love to help out! Check the code before merging, because I'm terrible at python. Pull request: #14

ArunMichaelDsouza commented 5 years ago

Thanks a lot @scheng123! Please view https://github.com/ArunMichaelDsouza/tensorflow-image-detection/pull/14#issuecomment-488709802