SyGoing
SyGoing
Thanks for your reply, I have implement the prelu with the pluginFactory while I didn't use the createPReLUPlugin ,since I don't know how to use it . I will read...
My PRelu implementation have not put on the github,while the other one have used the relu and scale method which is good for mtcnn's prelu layer: https://github.com/PKUZHOU/MTCNN_FaceDetection_TensorRT
哦哦 ,谢谢大佬,1)但是如果一个客户端服务器一次性带6路以上的摄像头进行检测,在加上其他的算法占用线程,会不会拖慢检测速度,2)原始mtcnn在监控视角的关键点检测受人头姿态影响,大佬的mtcnn关键点检测是否有提升呢? 后面我自己再测试一下关键点精度
哦哦,谢谢大佬,大角度问题这个我已经解决了,只是不是通过模型本身,所以问您这个问题的。非常感谢大佬,具体我多测试设计一下
the same question
那效果怎么样呢,有没有评价过呢,加入关键点是用的wilderface 吗(retinaface用的那个版本),?
我仅仅过滤掉了小于10的 我是先转换成voc再coco
如何开启 3399pro的ntb模式啊 大佬 @Amanda-Barbara
@Laughing-q 
def train_v8_detect(): overrides = {"data": "custom_boat.yaml", "resume":False, "model":"yolov8n.pt", "imgsz": 640, "epochs": 300, "save": True, "device":0, "batch":32, "pretrained": True, "v5loader":True } CFG.data = "custom_boat.yaml" # Trainer trainer = detect.DetectionTrainer(overrides=overrides) trainer.train() #...