zubairahmed-ai / live-background-blur Goto Github PK

Hello, Zubair!
I would like to use deeplab on macOS Python 3.7. I have to install tensorflow v1.15 (due to deeplab compatibility problem).
I install tensorflow models and extract it to:
./PROJECT_NAME/venv/lib/python3.7/site-packages/tensorflow.

Now, I have read in tensorflow installation guide, that I should execute in terminal:

>>> $ pwd
>>> ./PROJECT_NAME/venv/lib/python3.7/site-packages/tensorflow/models/research
>>> $ export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim.

After that I test it from the same path: $ python deeplab/model_test.py and it works fine:

Ran 5 tests in 35.463s
OK (skipped=1)

PROBLEM: So, I would like to test it in my project in path PROJECT_NAME/main.py simple way:

from matplotlib import pyplot as plt
import cv2

from model import Deeplabv3
# according your code I trying to run this
deeplab_model = Deeplabv3()

and $ python main.py gives me:

    from model import Deeplabv3
ModuleNotFoundError: No module named 'model'

I have already tried to move deeplab model folder to ./PROJECT_NAME/ and in another way move main.py to deeplab folder:

from deeplab import model # this way import
deeplab_model = Deeplabv3()
NameError: name 'Deeplabv3' is not defined

I'm not very familiar in python. Thanks for any information!

Hi, I tried your implementation and it works fine, it blurred the background pretty well, but I don't have the same performance you show in video in the README.md, Could there be a problem with the way I set up the project? I couldn't install cuda, so i set up the older version of keras-deeplab-v3-plus. Now I'm having like 1 frame per second, how many do you have? I now that the slowest part is when deeplab_model.predict is called but I don't understand where the method predict is.
Thanks in advance.

I'm running this on:

• Ubuntu 18.04
• Proccessor model: Intel(R) Core(TM) i7-7500U CPU @ 2.70GHz

It seems like there is extra padding around the foreground object. Any ideas on how to reduce padding (smoothen) to get a more accurate profile/outline???

Code:
https://pastebin.com/UNeNFbgm

Related post:?
https://www.pyimagesearch.com/2018/11/26/instance-segmentation-with-opencv/

A simple yet effective update to our instance segmentation pipeline would be to potentially:

Use morphological operations to increase the size of our mask
Apply a small amount of Gaussian blurring to the mask itself, helping smooth the mask
Scale the mask values to the range [0, 1]
Create an alpha layer using the scaled mask
Overlay the smoothed mask + person ROI on the blurred background

Alternatively, you could compute the contours of the mask itself and then apply contour approximation to help create a “more smoothed” mask.

@zubairahmed-ai Hi, I don't find the model and can't find the module of Deeplabv3 ?