- using tensorflow framework and opencv for video manipulations and Python 3.6.8
- extract frames from the video
- convert them to black and white [optional]
- recolor the frames using transfer learning on VGG16 [optional]
- run object masking with Mask RCNN
- collect the frames to a video
- model.py -- the main. specify the models to use. run it with video path as an argument
python model.py ./myVideo.mp4
- Colorize -- directory for recolor black and white models
- Framer -- directory to convert frames to video and vice versa
- Mask_RCNN -- directory of Mask RCNN model and code
- requirements.txt -- requirement for this project. NOTE: some requirement can't be installed with pip alone. there are comments in this file read them!
- better model for colorization, as vgg16 accept low resolution images
- faster detection algorithm to support live stection, maybe mot masking only detection
- improve preformance without scaling the GPU
- run model on a live video with minimal latency
- ???
for this demo i only wanted to look for:
- person, bicycle, car, motorcycle, bus, train, truck, traffic light, stop sign
- you can have much more labels
