Object Detection with YOLOV3
This is an Implementation of Object Detection using YOLO V3 trained on Pascal VOC Dataset with over 61 Million Parameters.
This model can detect only the objects listed in the above detection list.
DETECTION LIST = ['aeroplane', 'bicycle', 'bird', 'boat', 'bottle','bus', 'car', 'cat', 'chair', 'cow', 'diningtable','dog', 'horse', 'motorbike', 'person', 'pottedplant', 'sheep', 'sofa', 'train', 'tvmonitor']
NOTE : For Real-Time object detection make sure you have a good memory and Gpu on your computer system,a single forward pass takes on average 700-800 MB of memory.On a NVIDIA TITANX GPU this model can detect objects at 40-90 FPS which makes it very suited for Real-Time Object Detction.
=> Images are resized to 416x416 since the model accepts a fixed size image as input.
=> Each bounding box displays the object name it has detected and the probability of an object being present inside the bounding box.
#----------------------------------------------------------------------------------
STEPS TO USE THIS MODEL AS API :
Step 1] Install all required libraries and dependencies named below using Pip.
(torch , cv2 , numpy , matplotlib , tqdm , os , pandas , PIL)
Step 2] Download this repo and open a new project with the main file being main.py
Step 3] Download the pretrained weights required for the YoloV3 model from here
Step 4] The detect_objects( ) function in main.py acts as an interface to the model,pass the location of your image & weights file to the function & it'll plot back a new image with objects detected.
NOTE : The model sometimes produces INCORRECT PREDICTIONS & needs to be trained for more epochs to increase accuracy.
(GPU RECOMMENDED FOR RUNNING THE MODEL)
#----------------------------------------------------------------------------------
Original Repository : here (Great Job from the Author !)
Changes Made :
1] Created a Simple API interface of the model for easy usability by others.
2] Modified some code for faster preprocessing & postprocessing of images.