Object-Detection-using-Yolov5

Yolov5 Custom Dataset Training

A. Dataset Creation

collect images for custom dataset training.(size and resolution of the image dosent matter)
install python using https://www.python.org/downloads/
open command prompt and install lxml by typing "pip install lxml"
open command prompt and install PyQt5 by typing "pip install PyQt5" or "pip3 install PyQt5"
open labelImg folder in command prompt and type "python labelImg.py" for open labelImg.(labelImg is used for annotating the images)
in labelImg open collected image directory and start annotating the images using create rectbox tool and mention class name after that save all annotation in pascalvoc format.
after completing annotation copy all xml file to convert folder in \xml_to_yolo_convert\convert location.
then edit convert.py file.edit the class name in these lines with our class names(like "accessory,"top","rubber" etc the class name given in the time of annotation) and remove unwanted classes
save convert.py file
then open command prompt in that convert folder and type "python convert.py" this will create yolo text file(.txt) for the xml file

B. Seperating Dataset for train and validation

for training we need 80% of images we collected and for validation the remaining 20% image needed ( train=80% and validation=20% ).
copy 80% of images from dataset to train folder in location dataset\images\train
copy corresponding yolo file of the image used for train to location dataset\labels\train
copy remaining images from dataset to validation folder in location dataset\images\val
copy corresponding yolo file of the image used for validation to location dataset\labels\val

C. Yaml file editing

D. Yolov5 training and detecting

open command prompt in folder yolov5 then type "pip install -r requirements.txt" for installing requirement for yolov5
for training open command prompt in folder yolov5 then type "python train.py --img 640 --batch 16 --epochs 100 --data custom_dataset.yaml --weights yolov5s.pt --cache"
after training it will create weight in folder yolov5/runs/train/exp/weights/best.pt
for detecting images copy some images for testing in folder yolov5\data\images
open command prompt in folder yolov5 then type "python detect.py --source data/images/ --weights runs/train/exp/weights/best.pt"
after completing detection it gives results in folder yolov5/runs/detect

E. Keywords used in Yolov5

batch — batch size (-1 for auto batch size). Use the largest batch size that your hardware allows for.
epochs — number of epochs.
data — path to the data-configurations file.
cfg — path to the model-configurations file.
weights — path to initial weights.
cache — cache images for faster training.
img — image size in pixels (default — 640).
source — input path (0 for webcam)
conf — confidence threshold
iou — IoU threshold for NMS (Non Max Supression)
augment — augmented inference (TTA)

F. Create Virtual Environment for python ( optional )

download and install python
create a folder in our local disk
open newely created folder in command prompt
type "python -m venv virtual_environment_name" in the place of virtual_environment_name we can give our own virtual environment name by replacing it
type "virtual_environment_name\Scripts\activate" for activating the virtual environment
type "deactivate" for deactivating virtual environment
if pip not showing in virtual folder type "python -m ensurepip" then upgrade pip

G. Yolo Commands

python train.py --img 415 --batch 16 --epochs 30 --data dataset.yaml --weights yolov5s.pt --cache
python train.py --img 640 --batch 8 --epochs 100 --data mat.yaml --weights yolov5s.pt --cache
python detect.py --source data/images/ --weights runs/train/exp/weights/best.pt
python detect.py --source data/images/ --weights runs/train/exp/weights/best.pt --conf 0.50
python export.py --data mat.yaml --weights runs/train/exp/weights/best.pt --include tflite --img 640
python detect.py --weights runs/train/exp/weights/best-fp16.tflite --img 640 --source data/images/

H. References

minaahmed / object-detection-using-yolov5 Goto Github PK