search

pjreddie / darknet

Convolutional Neural Networks
http://pjreddie.com/darknet/
Other
25.85k stars 21.33k forks source link

FPS on YOLOv2-tiny, YOLOv3-tiny and YOLOv4-tiny ~same (CPU) #2232

Open bkhti4 opened 4 years ago

bkhti4 commented 4 years ago

Hello, I am using the tiny variants of YOLO on darknet_ros [https://github.com/tom13133/darknet_ros.git]. It seems that on all of them I am getting around 0.9-1.1 FPS on just CPU (core i7 if that is any helpful). My question is, does switching the tiny variants improves speed performances or gives slightly better detection accuracy.

Thanks

DiptanshuMalviya commented 4 years ago

hye bhakti4 , i am not answering your question, but i want to know had you done train model in yolov4_tiny ? if yes, please let me know how u had done... i am getting an error of load_detection thanks....

AliSheheryar commented 2 years ago

Bhakti4 ! I am running tiny yolo_v3 on ARM cpu ( No GPU) and inference time for single image is 1.01 sec. Is there any method for making it faster on CPU.???

bkhti4 commented 2 years ago

@AliSheheryar I don't know whether you only want the Yolov3 variant. Otherwise, you could convert to ONNX (for CPU using Yolov5, the method is linked here.

AliSheheryar commented 2 years ago

Thanks for letting me know. I really need to optimize for YOLO-v3 tiny. ARMNN only supports v3 Yolo last I checked. SO if there is anything v3 version can do to increase FPS.