toxtli / deepstab

Real-time Video Object Stabilization tool by using Deep Learning

DeepStab

Deepstab is a tool that is able to detect an specific object on a video and stabilize it by tracking the object and centering the image. The tool is compatible for models trained in TensorFlow Object Detection API, and Caffee models. There is a full explanation of how this works in this article this article.

Installation

Run

pip install -r requirements.txt

Quickstart

Example how to run it:

python app.py

In the built in example, the model is a face detection model (tf_ssd_mobilenet_openimages.pb) and you can tested it with your webcam by placing your face in front of it and moving around.

Models can be found here https://drive.google.com/drive/folders/1Wak-kieXQtpo8UtqL30WDFk1kfQVYCxu?usp=sharing

deepstab's Issues

error on running app.py

nvm

AttributeError: module 'cv2.cv2' has no attribute 'xfeatures2d'

So, I installed the requirements by running
pip install -r requirements.txt

But when I run it:
python app.py

I get the following error:

Traceback (most recent call last):
  File "app.py", line 185, in <module>
    surf = cv2.xfeatures2d.SURF_create()
AttributeError: module 'cv2.cv2' has no attribute 'xfeatures2d'

Any ideas? Thank you!

Port it to js.

Any plans to port to TensorFlow.js?