Yolov3 is an algorithm that uses deep convolutional neural networks to perform object detection. This repository implements Yolov3 using TensorFlow 2.0 and creates two easy-to-use APIs that you can integrate into web or mobile applications.
# Tensorflow CPU
conda env create -f conda-cpu.yml
conda activate yolov3-cpu
# Tensorflow GPU
conda env create -f conda-gpu.yml
conda activate yolov3-gpu# TensorFlow CPU
pip install -r requirements.txt
# TensorFlow GPU
pip install -r requirements-gpu.txt# Ubuntu 18.04
sudo apt-add-repository -r ppa:graphics-drivers/ppa
sudo apt install nvidia-driver-430
# Windows/Other
https://www.nvidia.com/Download/index.aspxFor Linux: Let's download official yolov3 weights pretrained on COCO dataset.
# yolov3
wget https://pjreddie.com/media/files/yolov3.weights -O weights/yolov3.weights
# yolov3-tiny
wget https://pjreddie.com/media/files/yolov3-tiny.weights -O weights/yolov3-tiny.weights
For Windows: You can download the yolov3 weights by clicking here and yolov3-tiny here then save them to the weights folder.
Learn How To Train Custom YOLOV3 Weights Here: https://www.youtube.com/watch?v=zJDUhGL26iU
Add your custom weights file to weights folder and your custom .names file into data/labels folder.
Load the weights using load_weights.py script. This will convert the yolov3 weights into TensorFlow .ckpt model files!
# yolov3
python load_weights.py
# yolov3-tiny
python load_weights.py --weights ./weights/yolov3-tiny.weights --output ./weights/yolov3-tiny.tf --tiny
After executing one of the above lines, you should see .tf files in your weights folder.
Now you can run a Flask application to create two object detections APIs in order to get detections through REST endpoints.
If you used custom weights and classes then you may need to adjust one or two of the following lines within the app.py file before running it.
You may also want to configure IOU threshold (how close two of the same class have to be in order to count it as one detection), the Confidence threshold (minimum detected confidence of a class in order to count it as a detection), or the maximum number of classes that can be detected in one image and all three can be adjusted within the yolov3-tf2/models.py file.
Initialize and run the Flask app on port 5000 of your local machine by running the following command from the root directory of this repo in a command prompt or shell.
python app.pyYou should see the following appear in the command prompt if the app is successfully running.
Detections API (http://localhost:5000/detections)
While app.py is running the first available API is a POST routed to /detections on port 5000 of localhost. This endpoint takes in images as input and returns a JSON response with all the detections found within each image (classes found within the images and the associated confidence)
You can test out the APIs using Postman or through Curl commands (both work fine). You may have to download them if you don't already have them.
Access the /detections API through Postman by doing the following.
Note that the body has to have key "images of type "form-data" set to file. When uploading files hold CTRL button and click to choose multiple photos.
The response should look similar to this.
To access and test the API through Curl, open a second command prompt or shell (may have to run as Administrator). Then cd your way to the root folder of this repository (Object-Detection-API) and run the following command.
curl.exe -X POST -F images=@data/images/dog.jpg "http://localhost:5000/detections"The JSON response should be outputted to the commmand prompt if it worked successfully.
Image API (http://localhost:5000/image)
While app.py is running the second available API is a POST routed to /image on port 5000 of localhost. This endpoint takes in a single image as input and returns a string encoded image as the response with all the detections now drawn on the image.
Access the /image API through Postman by configuring the following.
The uploaded image should be returned with the detections now drawn.
To access and test the API through Curl, open a second command prompt or shell (may have to run as Administrator). Then cd your way to the root folder of this repository (Object-Detection-API) and run the following command.
curl.exe -X POST -F images=@data/images/dog.jpg "http://localhost:5000/image" --output test.pngThis will save the returned image to the current folder as test.png (can't output the string encoded image to command prompt)
NOTE: As a backup both APIs save the images with the detections drawn overtop to the /detections folder upon each API request.
These are the two APIs I currently have created for Yolov3 Object Detection and I hope you find them useful. Feel free to integrate them into your applications as needed.
The tensorflow model can also be run not using the APIs but through using detect.py script.
Don't forget to set the IoU (Intersection over Union) and Confidence Thresholds within your yolov3-tf2/models.py file
Let's run an example or two using sample images found within the data/images folder.
# yolov3
python detect.py --images "data/images/dog.jpg, data/images/office.jpg"
# yolov3-tiny
python detect.py --weights ./weights/yolov3-tiny.tf --tiny --images "data/images/dog.jpg"
# webcam
python detect_video.py --video 0
# video file
python detect_video.py --video data/video/paris.mp4 --weights ./weights/yolov3-tiny.tf --tiny
# video file with output saved (can save webcam like this too)
python detect_video.py --video path_to_file.mp4 --output ./detections/output.aviThen you can find the detections in the detections folder.
You should see these two images saved for running the first command.
detection1.jpg
detection2.jpg
load_weights.py:
--output: path to output
(default: './weights/yolov3.tf')
--[no]tiny: yolov3 or yolov3-tiny
(default: 'false')
--weights: path to weights file
(default: './weights/yolov3.weights')
--num_classes: number of classes in the model
(default: '80')
(an integer)
detect.py:
--classes: path to classes file
(default: './data/labels/coco.names')
--images: path to input images as a string with images separated by ","
(default: 'data/images/dog.jpg')
--output: path to output folder
(default: './detections/')
--[no]tiny: yolov3 or yolov3-tiny
(default: 'false')
--weights: path to weights file
(default: './weights/yolov3.tf')
--num_classes: number of classes in the model
(default: '80')
(an integer)
detect_video.py:
--classes: path to classes file
(default: './data/labels/coco.names')
--video: path to input video (use 0 for webcam)
(default: './data/video/paris.mp4')
--output: path to output video (remember to set right codec for given format. e.g. XVID for .avi)
(default: None)
--output_format: codec used in VideoWriter when saving video to file
(default: 'XVID)
--[no]tiny: yolov3 or yolov3-tiny
(default: 'false')
--weights: path to weights file
(default: './weights/yolov3.tf')
--num_classes: number of classes in the model
(default: '80')
(an integer)