https://github.com/jwhittle933/tf-ibrith
Image classifier for עברית
https://github.com/jwhittle933/tf-ibrith
docker hebrew image-classification image-processing image-recognition python tensorflow
Last synced: about 2 months ago
JSON representation
Image classifier for עברית
- Host: GitHub
- URL: https://github.com/jwhittle933/tf-ibrith
- Owner: jwhittle933
- Created: 2019-06-15T14:43:16.000Z (almost 6 years ago)
- Default Branch: master
- Last Pushed: 2019-06-24T15:39:23.000Z (almost 6 years ago)
- Last Synced: 2025-02-02T18:33:07.808Z (4 months ago)
- Topics: docker, hebrew, image-classification, image-processing, image-recognition, python, tensorflow
- Language: Python
- Size: 493 MB
- Stars: 0
- Watchers: 3
- Forks: 0
- Open Issues: 4
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# עברית
## Training Hebrew Character Classifier
## Getting Started
The `src` directory houses the source code for the repo, but the scripts at the root can be used as a proxy to access them. `setup.sh` checks for the needed deps (tensorflow and numpy) which are used in `retrain.py` and `label_image.py` and installs them if needed. You don't need to call this script directly, as `train.sh` and `label.sh` call these before ever running the script. `graph.sh` opens `tensorboard` and accepts a single argument: the path to the graph. Graphs are kept in `src/graphs` the file names are saved in the format date|timestamp. This will be the practice for all stored files (graphs, output labels, saved models directories, and summaries).
The `modules` directory contains our labeled images. Images are labeled by the directory name that holds them, i.e., `aleph` dir holds images with label `aleph` for training. It's important that we don't nest any directories below this label level, as TF won't be able to train and will error out. As well, more than 1 directory must be present to train (as `retrain.py` takes files at random from the various directories to train on), and there can be no invalid directories, i.e., a directory named `mem` that is empty. Moreover, each image must be a .jpg/.jpeg, as TF will uses pixels to train on the images (opposed to a .png, which uses lines). If you screen grab a image from a Mac, the image will be saved as a png and need to be converted. Open the image and export as jpeg, and set the resolution to the highest level. It's also important to mmake sure our images are very high quality. The screen grabs currently in `modules` were grabbed from the [Digital Dead Sea Scrolls](http://dss.collections.imj.org.il/) online, and are very high quality. There is also the [Dead Sea Scrolls Archive](https://www.deadseascrolls.org.il/explore-the-archive?locale=en_US) for quality images. As well, there's lots of natural distortion in the images due to the decaying nature of the vellum/parchment/papyrus. TF can accept randomized image interference, but these images have some naturally. This is a great place to go grab more images if the modules need to be beefed up (and they definitely do). A good prelimary goal is to get at least __100__ images into each module, for all 22 Hebrew characters.
## Classifying
The `classify.go` script in the root is a command-line script for classifying an image. It has been adapted from [ActiveState/gococo](https://github.com/ActiveState/gococo), which was adapted from the [Tensorflow repo example](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/go/example_inception_inference_test.go). The script can be invoked (after install) as follows:
```go
tf_ibrith -dir= -jpg= [-out=< month-day-year|hour:min:sec.jpg>] [-labels=]
```
Or with `go run`
```go
go run classify.go -dir="MobileNet.06-24-2019|05:31:03" -jpg=heb-word.jpg
```
## Links
* [Build a Classifier](https://www.youtube.com/watch?v=QfNvhPx5Px8)
* [TF in 5 Minutes](https://www.youtube.com/watch?v=2FmcHiLCwTU)
* [Image Classifier](https://www.tensorflow.org/hub/tutorials/image_retraining)
* [Docker Container ](https://hub.docker.com/r/tensorflow/tensorflow/)
* [Image Labels](https://towardsdatascience.com/multi-label-image-classification-with-inception-net-cbb2ee538e30)
* [Object Detection](https://github.com/EdjeElectronics/TensorFlow-Object-Detection-API-Tutorial-Train-Multiple-Objects-Windows-10#1-install-tensorflow-gpu-15-skip-this-step-if-tensorflow-gpu-15-is-already-installed)
* [Golang Object Detection](https://github.com/ActiveState/gococo)