deepvision

A simple, high level, easy-to-use open source Computer Vision library for Python.

Installation

Installing dependencies

Provided the below python packages are installed, deepvision is completely pip installable.

• OpenCV
• TensorFlow

If you don't have them already installed, you can install through pip

pip install opencv-python tensorflow

Optional

or you can compile them from source if you want to enable optimizations for your specific hardware for better performance. If you are working with GPU, you can install tensorflow-gpu package through pip. Make sure you have the necessary Nvidia drivers installed preoperly (CUDA ToolKit, CuDNN etc).

If you are not sure, just go with the cpu-only tensorflow package.

You can also compile OpenCV from source to enable CUDA optimizations for Nvidia GPU.

Installing deepvision

pip install deepvision

To upgrade to the newest version pip install --upgrade deepvision

Optional

If you want to build deepvision from source, clone this repository and run the below commands.

git clone https://github.com/Lynchez/deepvision.git
cd deepvision
pip install .

Note: Compatability with Python 2.x is not officially tested.

Object detection

Detecting common objects in the scene is enabled through a single function call detect_common_objects(). It will return the bounding box co-ordinates, corrensponding labels and confidence scores for the detected objects in the image.

Example :

import deepvision as dv
from deepvision.object_detection import draw_bbox

bbox, label, conf = dv.detect_common_objects(img)

output_image = draw_bbox(img, bbox, label, conf)

Underneath it uses YOLOv4 model trained on COCO dataset capable of detecting 80 objects.

To enable GPU and Object Tracking

from deepvision.object_tracker import Track

boxes, classes, confidence = dv.detect_common_objects(frame, confidence=0.25, model='yolov4-tiny', enable_gpu=True)

frame = Track(boxes, classes, confidence, frame)

Checkout object_detection.py in examples directory for the complete code.

Object detection and Object tracking

YOLOv4 is actually a heavy model to run on CPU. If you are working with real time webcam / video feed and doesn't have GPU, try using tiny yolo which is a smaller version of the original YOLO model. It's significantly fast but less accurate.

bbox, label, conf = dv.detect_common_objects(img, confidence=0.25, model='yolov4-tiny')

Check out the example to learn more.

Other supported models: YOLOv3, YOLOv3-tiny.

Custom trained YOLO weights

To run inference with custom trained YOLOv3/v4 weights try the following

from deepvision.object_detection import YOLO

yolo = YOLO(weights, config, labels)
bbox, label, conf = yolo.detect_objects(img)
yolo.draw_bbox(img, bbox, label, conf)

To enable GPU

bbox, label, conf = yolo.detect_objects(img, enable_gpu=True)

Checkout the example to learn more.

Object detection with TFlite

To run inference with custom trained or pretrained weights from tensorflow hub and try the following

from deepvision import TFlite

tflite = TFlite("weights.tflite", "coconames.txt")
frame = tflite.detect_objects(frame)

Face detection

Detecting faces in an image is as simple as just calling the function detect_face(). It will return the bounding box corners and corresponding confidence for all the faces detected.

Example :

import deepvision as dv
faces, confidences = dv.detect_face(image)

Seriously, that's all it takes to do face detection with deepvision. Underneath it is using OpenCV's dnn module with a pre-trained caffemodel to detect faces.

To enable GPU

faces, confidences = dv.detect_face(image, enable_gpu=True)

Checkout face_detection.py in examples directory for the complete code.

Sample output :

Gender detection

Once face is detected, it can be passed on to detect_gender() function to recognize gender. It will return the labels (man, woman) and associated probabilities.

Example

label, confidence = dv.detect_gender(face)

Underneath deepvision is using an AlexNet-like model trained on Adience dataset by Gil Levi and Tal Hassner for their CVPR 2015 paper.

To enable GPU

label, confidence = dv.detect_gender(face, enable_gpu=True)

Checkout gender_detection.py in examples directory for the complete code.

Sample output :

Utils

Video to frames

get_frames( ) method can be helpful when you want to grab all the frames from a video. Just pass the path to the video, it will return all the frames in a list. Each frame in the list is a numpy array.

import deepvision as dv
frames = dv.get_frames('~/Downloads/demo.mp4')

Optionally you can pass in a directory path to save all the frames to disk.

frames = dv.get_frames('~/Downloads/demo.mp4', '~/Downloads/demo_frames/')

Creating gif

animate( ) method lets you create gif from a list of images. Just pass a list of images or path to a directory containing images and output gif name as arguments to the method, it will create a gif out of the images and save it to disk for you.

dv.animate(frames, '~/Documents/frames.gif')

Citation

If you find deepvision helpful in your work, please cite the following

@misc{ar2021deepvision,
  author =       {Nurettin Sinanoğlu},
  title =        {deepvision - Easy to use Computer Vision library for Python},
  howpublished = {\url{https://github.com/Lynchez/deepvision}},
  year =         {2021}
}