Performance

Model	Image Size	Epoch	Batch Size	Optimizer
YOLOv7	640	100 150 200	16	Adam
YOLOv7-X	640	100 150 200	8	Adam
YOLOv7-tiny	640	100 150 200	32	Adam

Training

YOLOv7 epoch (100, 150 200)

nama folder ('v100', 'v150', 'v200')

python train.py --cfg /cfg/training/yolov7.yaml --weights /weights/yolov7.pt --data /datasets/data.yaml --batch-size 8 --epochs [nilai epoch] --adam --name '[nama folder]'

YOLOv7x epoch (100, 150 200)

nama folder ('x100', 'x150', 'x200')

python train.py --cfg /cfg/training/yolov7x.yaml --weights /weights/yolov7x.pt --data /datasets/data.yaml --batch-size 16 --epochs [nilai epoch] --adam --name '[nama folder]'

YOLOv7-tiny epoch (100, 150 200)

nama folder ('t100', 't150', 't200')

python train.py --cfg /cfg/training/yolov7-tiny.yaml --weights /weights/yolov7-tiny.pt --data /datasets/data.yaml --batch-size 32 --epochs [nilai epoch] --adam --name '[nama folder]'

Deteksi

Camera Laptop

python detect.py --weights '[lokasi weights]' --source 0 --img 640

File gambar

python detect.py --source '[lokasi file]' --weights '[lokasi weights]' --img 640

Export

Pytorch to ONNX with NMS (and inference)

python export.py --weights yolov7-tiny.pt --grid --end2end --simplify \
        --topk-all 100 --iou-thres 0.65 --conf-thres 0.35 --img-size 640 640 --max-wh 640

Pytorch to TensorRT with NMS (and inference)

wget https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7-tiny.pt
python export.py --weights ./yolov7-tiny.pt --grid --end2end --simplify --topk-all 100 --iou-thres 0.65 --conf-thres 0.35 --img-size 640 640
git clone https://github.com/Linaom1214/tensorrt-python.git
python ./tensorrt-python/export.py -o yolov7-tiny.onnx -e yolov7-tiny-nms.trt -p fp16

Pytorch to TensorRT another way

wget https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7-tiny.pt
python export.py --weights yolov7-tiny.pt --grid --include-nms
git clone https://github.com/Linaom1214/tensorrt-python.git
python ./tensorrt-python/export.py -o yolov7-tiny.onnx -e yolov7-tiny-nms.trt -p fp16

# Or use trtexec to convert ONNX to TensorRT engine
/usr/src/tensorrt/bin/trtexec --onnx=yolov7-tiny.onnx --saveEngine=yolov7-tiny-nms.trt --fp16

Tested with: Python 3.7.13, Pytorch 1.12.0+cu113