trt multi-thread
This commit is contained in:
dell
parent
2eccbf8c7c
commit
e74ad42690
2
demo.py
2
demo.py
|
|
@ -18,7 +18,7 @@ def river_demo_v2():
|
|||
###注意TRT模型生成时，就需要对应cuda device，下面的trt文件是cuda:0生成的，device只能是0
|
||||
##以下参数目前不可改
|
||||
|
||||
gpuname='3090'
|
||||
gpuname='3090'###3090,2080Ti. 只能是其中之一
|
||||
Detweights = "../AIlib/weights/yolov5/class8/bestcao_%s_fp16.engine"%(gpuname) ##升级后的检测模型
|
||||
labelnames = "../AIlib/weights/yolov5/class8/labelnames.json" ##对应类别表
|
||||
|
||||
|
|
|
|||
|
|
@ -0,0 +1,381 @@
|
|||
import sys
|
||||
from concurrent.futures import ThreadPoolExecutor
|
||||
|
||||
sys.path.extend(['..','../AIlib' ])
|
||||
|
||||
from AI import AI_process,AI_process_v2,AI_process_forest,get_postProcess_para
|
||||
import cv2,os,time
|
||||
from segutils.segmodel import SegModel
|
||||
from segutils.segmodel_trt import SegModel_STDC_trt
|
||||
from segutils.trtUtils import DetectMultiBackend
|
||||
from models.experimental import attempt_load
|
||||
from utils.torch_utils import select_device
|
||||
from utilsK.queRiver import get_labelnames,get_label_arrays
|
||||
import numpy as np
|
||||
import torch
|
||||
import tensorrt as trt
|
||||
from utilsK.masterUtils import get_needed_objectsIndex
|
||||
|
||||
'''
|
||||
多线程
|
||||
'''
|
||||
|
||||
def process_v1(frame):
|
||||
#try:
|
||||
|
||||
time00 = time.time()
|
||||
H,W,C = frame[0][0].shape
|
||||
p_result,timeOut = AI_process(frame[0],frame[1],frame[2],frame[3],frame[4],frame[5],frame[6],frame[7],frame[8], frame[9],frame[10],font=frame[11],trtFlag_det=frame[13],trtFlag_seg=frame[14],segPar=frame[15])
|
||||
time11 = time.time()
|
||||
image_array = p_result[1]
|
||||
#cv2.imwrite(os.path.join('images/results/',frame[12] ) ,image_array)
|
||||
time22 = time.time()
|
||||
print('%s,%d*%d,AI-process: %.1f,image save:%.1f , %s'%(frame[12],H,W, (time11 - time00) * 1000.0, (time22-time11)*1000.0,timeOut))
|
||||
return 'success'
|
||||
#except Exception as e:
|
||||
# return 'failed:'+str(e)
|
||||
|
||||
def river_demo_v3():
|
||||
##预先设置的参数
|
||||
device_='0' ##选定模型，可选 cpu,'0','1'
|
||||
###注意TRT模型生成时，就需要对应cuda device，下面的trt文件是cuda:0生成的，device只能是0
|
||||
##以下参数目前不可改
|
||||
labelnames = "../AIlib/weights/yolov5/class8/labelnames.json" ##对应类别表
|
||||
gpuname='3090';
|
||||
max_workers=6;
|
||||
trtFlag_det=True;trtFlag_seg=True
|
||||
device = select_device(device_)
|
||||
names=get_labelnames(labelnames)
|
||||
|
||||
half = device.type != 'cpu' # half precision only supported on CUDA
|
||||
if trtFlag_det:
|
||||
Detweights = "../AIlib/weights/yolov5/class8/bestcao_%s_fp16.engine"%(gpuname) ##升级后的检测模型
|
||||
|
||||
logger = trt.Logger(trt.Logger.INFO)
|
||||
with open(Detweights, "rb") as f, trt.Runtime(logger) as runtime:
|
||||
model=runtime.deserialize_cuda_engine(f.read())# 输入trt本地文件，返回ICudaEngine对象
|
||||
print('############locad det model trt success#######')
|
||||
else:
|
||||
Detweights = "../AIlib/weights/yolov5/class8/bestcao.pt"
|
||||
model = attempt_load(Detweights, map_location=device) # load FP32 model
|
||||
print('############locad det model pth success#######')
|
||||
if half: model.half()
|
||||
|
||||
|
||||
seg_nclass = 2
|
||||
#Segweights = "../AIlib/weights/BiSeNet/checkpoint.pth"
|
||||
segPar={'modelSize':(640,360),'mean':(0.485, 0.456, 0.406),'std' :(0.229, 0.224, 0.225),'numpy':False, 'RGB_convert_first':True}
|
||||
if trtFlag_seg:
|
||||
Segweights = '../AIlib/weights/STDC/model_maxmIOU75_1720_0.946_360640_%s_fp16.engine'%(gpuname) ##升级的分割模型
|
||||
segmodel = SegModel_STDC_trt(weights=Segweights,modelsize=(640,360),std=(0.229, 0.224, 0.225),mean=(0.485, 0.456, 0.406),device=device)
|
||||
|
||||
logger = trt.Logger(trt.Logger.INFO)
|
||||
with open(Segweights, "rb") as f, trt.Runtime(logger) as runtime:
|
||||
segmodel=runtime.deserialize_cuda_engine(f.read())# 输入trt本地文件，返回ICudaEngine对象
|
||||
print('############locad seg model trt success#######')
|
||||
else:
|
||||
Segweights = '../AIlib/weights/STDC/model_maxmIOU75_1720_0.946_360640.pth'
|
||||
segmodel = SegModel(nclass=seg_nclass,weights=Segweights,device=device)
|
||||
print('############locad seg model pth success#######')
|
||||
|
||||
postFile= '../AIlib/conf/para.json'
|
||||
digitFont= { 'line_thickness':2,'boxLine_thickness':1, 'fontSize':1.0,'waterLineColor':(0,255,255),'waterLineWidth':3}
|
||||
conf_thres,iou_thres,classes,rainbows=get_postProcess_para(postFile)
|
||||
|
||||
####模型选择参数用如下：
|
||||
mode_paras=[
|
||||
{"id":"0","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"1","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"2","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"3","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"4","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"5","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"6","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"7","config":{"k1":"v1","k2":"v2"}},
|
||||
]
|
||||
allowedList,allowedList_string=get_needed_objectsIndex(mode_paras)
|
||||
#allowedList=[0,1,2,3]
|
||||
##加载模型，准备好显示字符
|
||||
label_arraylist = get_label_arrays(names,rainbows,outfontsize=40,fontpath="../AIlib/conf/platech.ttf")
|
||||
|
||||
|
||||
##图像测试
|
||||
impth = 'images/slope/'
|
||||
outpth = 'images/results/'
|
||||
folders = os.listdir(impth)
|
||||
frames=[]
|
||||
|
||||
for i in range(len(folders)):
|
||||
imgpath = os.path.join(impth, folders[i])
|
||||
im0s=[cv2.imread(imgpath)]
|
||||
frame=(im0s,model,segmodel,names,label_arraylist,rainbows,half,device,conf_thres, iou_thres,allowedList,digitFont,folders[i],trtFlag_det,trtFlag_seg,segPar)
|
||||
frames.append(frame)
|
||||
t1=time.time()
|
||||
if max_workers==1:
|
||||
for i in range(len(folders)):
|
||||
t5=time.time()
|
||||
process_v1(frames[i])
|
||||
t6=time.time()
|
||||
print('#######%s, ms:%.1f , accumetate time:%.1f, avage:%1.f '%(folders[i], (t6-t5)*1000.0,(t6-t1)*1000.0, (t6-t1)*1000.0/(i+1)))
|
||||
else:
|
||||
with ThreadPoolExecutor(max_workers=max_workers) as t:
|
||||
for result in t.map(process_v1, frames):
|
||||
#print(result)
|
||||
t=result
|
||||
|
||||
t2=time.time()
|
||||
print('All %d images time:%.1f ms ,each:%.1f ms, with %d threads'%(len(folders),(t2-t1)*1000, (t2-t1)*1000.0/len(folders) , max_workers) )
|
||||
|
||||
def river_demo_v2():
|
||||
##预先设置的参数
|
||||
device_='0' ##选定模型，可选 cpu,'0','1'
|
||||
###注意TRT模型生成时，就需要对应cuda device，下面的trt文件是cuda:0生成的，device只能是0
|
||||
##以下参数目前不可改
|
||||
|
||||
trtFlag=False
|
||||
gpuname='3090'
|
||||
if trtFlag:
|
||||
Detweights = "../AIlib/weights/yolov5/class8/bestcao_%s_fp16.engine"%(gpuname) ##升级后的检测模型
|
||||
else:
|
||||
Detweights = "../AIlib/weights/yolov5/class8/bestcao.pt" ##升级后的检测模型
|
||||
|
||||
labelnames = "../AIlib/weights/yolov5/class8/labelnames.json" ##对应类别表
|
||||
|
||||
|
||||
seg_nclass = 2
|
||||
#Segweights = "../AIlib/weights/BiSeNet/checkpoint.pth"
|
||||
if trtFlag:
|
||||
Segweights = '../AIlib/weights/STDC/model_maxmIOU75_1720_0.946_360640_%s_fp16.engine'%(gpuname) ##升级的分割模型
|
||||
else:
|
||||
Segweights = '../AIlib/weights/STDC/model_maxmIOU75_1720_0.946_360640.pth'
|
||||
|
||||
postFile= '../AIlib/conf/para.json'
|
||||
digitFont= { 'line_thickness':2,'boxLine_thickness':1, 'fontSize':1.0,'waterLineColor':(0,255,255),'waterLineWidth':3}
|
||||
conf_thres,iou_thres,classes,rainbows=get_postProcess_para(postFile)
|
||||
|
||||
####模型选择参数用如下：
|
||||
mode_paras=[
|
||||
{"id":"0","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"1","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"2","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"3","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"4","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"5","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"6","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"7","config":{"k1":"v1","k2":"v2"}},
|
||||
]
|
||||
|
||||
allowedList,allowedList_string=get_needed_objectsIndex(mode_paras)
|
||||
#allowedList=[0,1,2,3]
|
||||
##加载模型，准备好显示字符
|
||||
device = select_device(device_)
|
||||
names=get_labelnames(labelnames)
|
||||
|
||||
label_arraylist = get_label_arrays(names,rainbows,outfontsize=40,fontpath="../AIlib/conf/platech.ttf")
|
||||
half = device.type != 'cpu' # half precision only supported on CUDA
|
||||
|
||||
if trtFlag:
|
||||
segmodel = SegModel_STDC_trt(weights=Segweights,modelsize=(640,360),std=(0.229, 0.224, 0.225),mean=(0.485, 0.456, 0.406),device=device)
|
||||
model = DetectMultiBackend(Detweights, device=device)
|
||||
else:
|
||||
model = attempt_load(Detweights, map_location=device) # load FP32 model
|
||||
if half: model.half()
|
||||
segmodel = SegModel(nclass=seg_nclass,weights=Segweights,device=device)
|
||||
|
||||
##图像测试
|
||||
#url='images/examples/20220624_响水河_12300_1621.jpg'
|
||||
impth = 'images/slope/'
|
||||
outpth = 'images/results/'
|
||||
folders = os.listdir(impth)
|
||||
for i in range(len(folders)):
|
||||
imgpath = os.path.join(impth, folders[i])
|
||||
im0s=[cv2.imread(imgpath)]
|
||||
H,W,C = im0s[0].shape
|
||||
time00 = time.time()
|
||||
p_result,timeOut = AI_process_v2(im0s,model,segmodel,names,label_arraylist,rainbows,half,device,conf_thres, iou_thres,allowedList,font=digitFont)
|
||||
time11 = time.time()
|
||||
image_array = p_result[1]
|
||||
cv2.imwrite( os.path.join( outpth,folders[i] ) ,image_array )
|
||||
print('%s,%d*%d,AI-process: %.1f, %s'%(folders[i],H,W, (time11 - time00) * 1000,timeOut))
|
||||
|
||||
def river_demo():
|
||||
##预先设置的参数
|
||||
device_='1' ##选定模型，可选 cpu,'0','1'
|
||||
|
||||
##以下参数目前不可改
|
||||
#Detweights = "../AIlib/weights/yolov5/class5/best_5classes.pt"
|
||||
#labelnames = "../AIlib/weights/yolov5/class5/labelnames.json"
|
||||
|
||||
Detweights = "../AIlib/weights/yolov5/class8/bestcao.pt" ##升级后的检测模型
|
||||
labelnames = "../AIlib/weights/yolov5/class8/labelnames.json" ##对应类别表
|
||||
|
||||
seg_nclass = 2
|
||||
#Segweights = "../AIlib/weights/BiSeNet/checkpoint.pth"
|
||||
Segweights = '../AIlib/weights/STDC/model_maxmIOU75_1720_0.946_360640.pth' ##升级的分割模型
|
||||
|
||||
postFile= '../AIlib/conf/para.json'
|
||||
digitFont= { 'line_thickness':2,'boxLine_thickness':1, 'fontSize':1.0,'waterLineColor':(0,255,255),'waterLineWidth':3}
|
||||
conf_thres,iou_thres,classes,rainbows=get_postProcess_para(postFile)
|
||||
|
||||
####模型选择参数用如下：
|
||||
mode_paras=[
|
||||
{"id":"0","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"1","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"2","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"3","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"4","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"5","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"6","config":{"k1":"v1","k2":"v2"}},
|
||||
{"id":"7","config":{"k1":"v1","k2":"v2"}},
|
||||
]
|
||||
|
||||
allowedList,allowedList_string=get_needed_objectsIndex(mode_paras)
|
||||
#allowedList=[0,1,2,3]
|
||||
|
||||
|
||||
|
||||
##加载模型，准备好显示字符
|
||||
device = select_device(device_)
|
||||
names=get_labelnames(labelnames)
|
||||
|
||||
label_arraylist = get_label_arrays(names,rainbows,outfontsize=40,fontpath="../AIlib/conf/platech.ttf")
|
||||
half = device.type != 'cpu' # half precision only supported on CUDA
|
||||
model = attempt_load(Detweights, map_location=device) # load FP32 model
|
||||
if half: model.half()
|
||||
|
||||
segmodel = SegModel(nclass=seg_nclass,weights=Segweights,device=device)
|
||||
|
||||
|
||||
|
||||
##图像测试
|
||||
#url='images/examples/20220624_响水河_12300_1621.jpg'
|
||||
impth = 'images/slope/'
|
||||
outpth = 'images/results/'
|
||||
folders = os.listdir(impth)
|
||||
for i in range(len(folders)):
|
||||
imgpath = os.path.join(impth, folders[i])
|
||||
im0s=[cv2.imread(imgpath)]
|
||||
H,W,C = im0s[0].shape
|
||||
time00 = time.time()
|
||||
p_result,timeOut = AI_process(im0s,model,segmodel,names,label_arraylist,rainbows,half,device,conf_thres, iou_thres,allowedList,font=digitFont)
|
||||
time11 = time.time()
|
||||
image_array = p_result[1]
|
||||
cv2.imwrite( os.path.join( outpth,folders[i] ) ,image_array )
|
||||
print('%s,%d*%d,AI-process: %.1f, %s'%(folders[i],H,W, (time11 - time00) * 1000,timeOut))
|
||||
|
||||
def road_forest_demo(business ):
|
||||
|
||||
##使用森林，道路模型,business 控制['forest','road']
|
||||
##预先设置的参数
|
||||
device_='1' ##选定模型，可选 cpu,'0','1'
|
||||
|
||||
|
||||
|
||||
##以下参数目前不可改
|
||||
#business='forest';imageW=4916 ####森林模型
|
||||
#business='road';
|
||||
imageW=1536 ####道路模型
|
||||
digitFont= { 'line_thickness':2, 'fontSize':1.0} ###数字显示的线宽度，大小; 如果都是None，则采用默认大小
|
||||
Detweights="../AIlib/weights/%s/best.pt"%(business)
|
||||
labelnames = "../AIlib/weights/%s/labelnames.json"%(business)
|
||||
|
||||
postFile= '../AIlib/conf/para.json'
|
||||
print( Detweights,labelnames )
|
||||
conf_thres,iou_thres,classes,rainbows=get_postProcess_para(postFile)
|
||||
|
||||
####模型选择参数用如下：
|
||||
mode_paras=[
|
||||
{
|
||||
"id":"0",
|
||||
"config":{
|
||||
"k1":"v1",
|
||||
"k2":"v2"
|
||||
}
|
||||
},
|
||||
{
|
||||
"id":"1",
|
||||
"config":{
|
||||
"k1":"v1",
|
||||
"k2":"v2"
|
||||
}
|
||||
}
|
||||
]
|
||||
|
||||
allowedList,allowedList_string=get_needed_objectsIndex(mode_paras)
|
||||
#allowedList=[0,1,2,3]
|
||||
|
||||
print('####line108###')
|
||||
|
||||
##只加载检测模型，准备好显示字符
|
||||
device = select_device(device_)
|
||||
names=get_labelnames(labelnames)
|
||||
#imageW=4915;###默认是1920，在森林巡检的高清图像中是4920
|
||||
outfontsize=int(imageW/1920*40);###
|
||||
label_arraylist = get_label_arrays(names,rainbows,outfontsize=outfontsize,fontpath="../AIlib/conf/platech.ttf")
|
||||
half = device.type != 'cpu' # half precision only supported on CUDA
|
||||
model = attempt_load(Detweights, map_location=device) # load FP32 model
|
||||
if half: model.half()
|
||||
segmodel = None
|
||||
|
||||
|
||||
|
||||
##图像测试
|
||||
#url='images/examples/20220624_响水河_12300_1621.jpg'
|
||||
impth = 'images/%s/'%(business)
|
||||
outpth = 'images/results/'
|
||||
folders = os.listdir(impth)
|
||||
folders.sort()
|
||||
for i in range(len(folders)):
|
||||
#for i in range(2):
|
||||
imgpath = os.path.join(impth, folders[i])
|
||||
im0s=[cv2.imread(imgpath)]
|
||||
time00 = time.time()
|
||||
#使用不同的函数。每一个领域采用一个函数
|
||||
p_result,timeOut = AI_process_forest(im0s,model,segmodel,names,label_arraylist,rainbows,half,device,conf_thres, iou_thres,allowedList,font=digitFont)
|
||||
time11 = time.time()
|
||||
image_array = p_result[1]
|
||||
cv2.imwrite( os.path.join( outpth,folders[i] ) ,image_array )
|
||||
|
||||
print('----image:%s, process:%s ,save:%s, %s'%(folders[i],(time11-time00) * 1000, (time.time() - time11) * 1000,timeOut) )
|
||||
|
||||
|
||||
def jkm_demo():
|
||||
from utilsK.jkmUtils import pre_process,post_process,get_return_data
|
||||
img_type = 'plate' ## code,plate
|
||||
par={'code':{'weights':'../AIlib/weights/jkm/health_yolov5s_v3.jit','img_type':'code','nc':10 },
|
||||
'plate':{'weights':'../AIlib/weights/jkm/plate_yolov5s_v3.jit','img_type':'plate','nc':1 },
|
||||
'conf_thres': 0.4,
|
||||
'iou_thres':0.45,
|
||||
'device':'cuda:0',
|
||||
'plate_dilate':(0.5,0.1)
|
||||
}
|
||||
###加载模型
|
||||
device = torch.device(par['device'])
|
||||
jit_weights = par['code']['weights']
|
||||
model = torch.jit.load(jit_weights)
|
||||
|
||||
jit_weights = par['plate']['weights']
|
||||
model_plate = torch.jit.load(jit_weights)
|
||||
|
||||
imgd='images/plate'
|
||||
imgpaths = os.listdir(imgd)
|
||||
for imgp in imgpaths[0:]:
|
||||
#imgp = 'plate_IMG_20221030_100612.jpg'
|
||||
imgpath = os.path.join(imgd,imgp)
|
||||
im0 = cv2.imread(imgpath) #读取数据
|
||||
img ,padInfos = pre_process(im0,device) ##预处理
|
||||
if img_type=='code': pred = model(img) ##模型推理
|
||||
else: pred = model_plate(img)
|
||||
boxes = post_process(pred,padInfos,device,conf_thres= par['conf_thres'], iou_thres= par['iou_thres'],nc=par[img_type]['nc']) #后处理
|
||||
dataBack=get_return_data(im0,boxes,modelType=img_type,plate_dilate=par['plate_dilate'])
|
||||
print(imgp,boxes,dataBack['type'])
|
||||
for key in dataBack.keys():
|
||||
if isinstance(dataBack[key],list):
|
||||
cv2.imwrite( 'images/results/%s_%s.jpg'%( imgp.replace('.jpg','').replace('.png',''),key),dataBack[key][0] ) ###返回值： dataBack
|
||||
|
||||
1
|
||||
if __name__=="__main__":
|
||||
river_demo_v3()
|
||||
#road_forest_demo('forest' )
|
||||
#jkm_demo()
|
||||
|
||||
Loading…
Reference in New Issue