如何测量得到论文中的检测与实例分割的延迟？

[QWERDF007]

我在i9-9900k和3070上测试得到的延迟是50ms左右. 测试代码：

# Copyright (c) OpenMMLab. All rights reserved.
import asyncio
from argparse import ArgumentParser

from mmdet.apis import (async_inference_detector, inference_detector,
                        init_detector, show_result_pyplot)

import repvit
import torch

from pathlib import Path
import time
from PIL import Image
import magic

def parse_args():
    parser = ArgumentParser()
    parser.add_argument('img', help='Image file')
    parser.add_argument('config', help='Config file')
    parser.add_argument('checkpoint', help='Checkpoint file')
    parser.add_argument('--out-file', default=None, help='Path to output file')
    parser.add_argument(
        '--device', default='cuda:0', help='Device used for inference')
    parser.add_argument(
        '--palette',
        default='coco',
        choices=['coco', 'voc', 'citys', 'random'],
        help='Color palette used for visualization')
    parser.add_argument(
        '--score-thr', type=float, default=0.3, help='bbox score threshold')
    parser.add_argument(
        '--async-test',
        action='store_true',
        help='whether to set async options for async inference.')
    args = parser.parse_args()
    return args

def main(args):
    # build the model from a config file and a checkpoint file
    model = init_detector(args.config, args.checkpoint, device=args.device)

    data_dir = Path(args.img)
    for fpath in data_dir.iterdir():
    # fpaths = [str(fpath) for fpath in data_dir.iterdir()]

        start = time.time()
        # test a single image
        result = inference_detector(model,  str(fpath))
        torch.cuda.synchronize()
        end = time.time()

        # print(f"Inference time: {end-start},  total: {len(fpaths)} average time: {(end-start)/len(fpaths)}")
        img = Image.open(str(fpath))

        print(f"{fpath} size: {img.width}x{img.height}, time: {end-start}")
        img.close()
    # show the results
    # show_result_pyplot(
    #     model,
    #     str(fpath),
    #     result,
    #     palette=args.palette,
    #     score_thr=args.score_thr,
    #     out_file=args.out_file)

async def async_main(args):
    # build the model from a config file and a checkpoint file
    model = init_detector(args.config, args.checkpoint, device=args.device)
    # test a single image
    tasks = asyncio.create_task(async_inference_detector(model, args.img))
    result = await asyncio.gather(tasks)
    # show the results
    show_result_pyplot(
        model,
        args.img,
        result[0],
        palette=args.palette,
        score_thr=args.score_thr,
        out_file=args.out_file)

if __name__ == '__main__':
    args = parse_args()
    if args.async_test:
        asyncio.run(async_main(args))
    else:
        main(args)

测试结果：

/data/wt/TestData/dataset/dogsvscats/test1/1.jpg size: 381x499, time: 0.08272171020507812
/data/wt/TestData/dataset/dogsvscats/test1/10.jpg size: 500x374, time: 0.06478381156921387
/data/wt/TestData/dataset/dogsvscats/test1/100.jpg size: 499x343, time: 0.07163786888122559
/data/wt/TestData/dataset/dogsvscats/test1/1000.jpg size: 288x499, time: 0.0686182975769043
/data/wt/TestData/dataset/dogsvscats/test1/10000.jpg size: 467x499, time: 0.048500776290893555
/data/wt/TestData/dataset/dogsvscats/test1/10001.jpg size: 499x293, time: 0.06814789772033691
/data/wt/TestData/dataset/dogsvscats/test1/10002.jpg size: 500x374, time: 0.06205487251281738
/data/wt/TestData/dataset/dogsvscats/test1/10003.jpg size: 500x263, time: 0.05933856964111328
/data/wt/TestData/dataset/dogsvscats/test1/10004.jpg size: 488x399, time: 0.05417633056640625
/data/wt/TestData/dataset/dogsvscats/test1/10005.jpg size: 249x227, time: 0.04686784744262695
/data/wt/TestData/dataset/dogsvscats/test1/10006.jpg size: 499x375, time: 0.05886483192443848
/data/wt/TestData/dataset/dogsvscats/test1/10007.jpg size: 500x374, time: 0.05744314193725586
/data/wt/TestData/dataset/dogsvscats/test1/10008.jpg size: 499x375, time: 0.05886197090148926
/data/wt/TestData/dataset/dogsvscats/test1/10009.jpg size: 489x367, time: 0.057126522064208984
/data/wt/TestData/dataset/dogsvscats/test1/1001.jpg size: 250x243, time: 0.045355796813964844
/data/wt/TestData/dataset/dogsvscats/test1/10010.jpg size: 500x374, time: 0.058531999588012695
/data/wt/TestData/dataset/dogsvscats/test1/10011.jpg size: 400x345, time: 0.05124783515930176
/data/wt/TestData/dataset/dogsvscats/test1/10012.jpg size: 335x409, time: 0.05216383934020996
/data/wt/TestData/dataset/dogsvscats/test1/10013.jpg size: 470x500, time: 0.04745936393737793
/data/wt/TestData/dataset/dogsvscats/test1/10015.jpg size: 500x387, time: 0.05611777305603027
/data/wt/TestData/dataset/dogsvscats/test1/10016.jpg size: 240x269, time: 0.049646854400634766
/data/wt/TestData/dataset/dogsvscats/test1/10017.jpg size: 250x293, time: 0.05157661437988281
/data/wt/TestData/dataset/dogsvscats/test1/10018.jpg size: 434x375, time: 0.05177617073059082
/data/wt/TestData/dataset/dogsvscats/test1/10019.jpg size: 499x432, time: 0.051323652267456055
/data/wt/TestData/dataset/dogsvscats/test1/1002.jpg size: 374x500, time: 0.05673623085021973
/data/wt/TestData/dataset/dogsvscats/test1/10020.jpg size: 199x200, time: 0.0448911190032959
/data/wt/TestData/dataset/dogsvscats/test1/10021.jpg size: 396x260, time: 0.06324148178100586
/data/wt/TestData/dataset/dogsvscats/test1/10022.jpg size: 500x374, time: 0.057634830474853516
/data/wt/TestData/dataset/dogsvscats/test1/10023.jpg size: 360x269, time: 0.05904746055603027
/data/wt/TestData/dataset/dogsvscats/test1/10024.jpg size: 303x359, time: 0.050252437591552734
/data/wt/TestData/dataset/dogsvscats/test1/10025.jpg size: 443x459, time: 0.04602837562561035
/data/wt/TestData/dataset/dogsvscats/test1/10026.jpg size: 499x375, time: 0.05867409706115723
/data/wt/TestData/dataset/dogsvscats/test1/10027.jpg size: 320x431, time: 0.05827951431274414
/data/wt/TestData/dataset/dogsvscats/test1/10028.jpg size: 200x149, time: 0.05838799476623535
/data/wt/TestData/dataset/dogsvscats/test1/10029.jpg size: 319x414, time: 0.05730867385864258
/data/wt/TestData/dataset/dogsvscats/test1/1003.jpg size: 499x398, time: 0.055377960205078125
/data/wt/TestData/dataset/dogsvscats/test1/10030.jpg size: 479x401, time: 0.05260920524597168
/data/wt/TestData/dataset/dogsvscats/test1/10031.jpg size: 274x500, time: 0.06286954879760742
/data/wt/TestData/dataset/dogsvscats/test1/10033.jpg size: 499x375, time: 0.06049299240112305
/data/wt/TestData/dataset/dogsvscats/test1/10034.jpg size: 406x400, time: 0.04815244674682617
/data/wt/TestData/dataset/dogsvscats/test1/10035.jpg size: 310x458, time: 0.061447858810424805
/data/wt/TestData/dataset/dogsvscats/test1/10036.jpg size: 499x375, time: 0.0570530891418457
/data/wt/TestData/dataset/dogsvscats/test1/10037.jpg size: 499x375, time: 0.06384038925170898
/data/wt/TestData/dataset/dogsvscats/test1/10038.jpg size: 499x466, time: 0.04932117462158203

[hellohahaw]

我在i9-9900k和3070上测试得到的延迟是50ms左右. 测试代码：

# Copyright (c) OpenMMLab. All rights reserved.
import asyncio
from argparse import ArgumentParser

from mmdet.apis import (async_inference_detector, inference_detector,
                        init_detector, show_result_pyplot)

import repvit
import torch

from pathlib import Path
import time
from PIL import Image
import magic

def parse_args():
    parser = ArgumentParser()
    parser.add_argument('img', help='Image file')
    parser.add_argument('config', help='Config file')
    parser.add_argument('checkpoint', help='Checkpoint file')
    parser.add_argument('--out-file', default=None, help='Path to output file')
    parser.add_argument(
        '--device', default='cuda:0', help='Device used for inference')
    parser.add_argument(
        '--palette',
        default='coco',
        choices=['coco', 'voc', 'citys', 'random'],
        help='Color palette used for visualization')
    parser.add_argument(
        '--score-thr', type=float, default=0.3, help='bbox score threshold')
    parser.add_argument(
        '--async-test',
        action='store_true',
        help='whether to set async options for async inference.')
    args = parser.parse_args()
    return args

def main(args):
    # build the model from a config file and a checkpoint file
    model = init_detector(args.config, args.checkpoint, device=args.device)

    data_dir = Path(args.img)
    for fpath in data_dir.iterdir():
    # fpaths = [str(fpath) for fpath in data_dir.iterdir()]

        start = time.time()
        # test a single image
        result = inference_detector(model,  str(fpath))
        torch.cuda.synchronize()
        end = time.time()

        # print(f"Inference time: {end-start},  total: {len(fpaths)} average time: {(end-start)/len(fpaths)}")
        img = Image.open(str(fpath))

        print(f"{fpath} size: {img.width}x{img.height}, time: {end-start}")
        img.close()
    # show the results
    # show_result_pyplot(
    #     model,
    #     str(fpath),
    #     result,
    #     palette=args.palette,
    #     score_thr=args.score_thr,
    #     out_file=args.out_file)

async def async_main(args):
    # build the model from a config file and a checkpoint file
    model = init_detector(args.config, args.checkpoint, device=args.device)
    # test a single image
    tasks = asyncio.create_task(async_inference_detector(model, args.img))
    result = await asyncio.gather(tasks)
    # show the results
    show_result_pyplot(
        model,
        args.img,
        result[0],
        palette=args.palette,
        score_thr=args.score_thr,
        out_file=args.out_file)

if __name__ == '__main__':
    args = parse_args()
    if args.async_test:
        asyncio.run(async_main(args))
    else:
        main(args)

测试结果：

/data/wt/TestData/dataset/dogsvscats/test1/1.jpg size: 381x499, time: 0.08272171020507812
/data/wt/TestData/dataset/dogsvscats/test1/10.jpg size: 500x374, time: 0.06478381156921387
/data/wt/TestData/dataset/dogsvscats/test1/100.jpg size: 499x343, time: 0.07163786888122559
/data/wt/TestData/dataset/dogsvscats/test1/1000.jpg size: 288x499, time: 0.0686182975769043
/data/wt/TestData/dataset/dogsvscats/test1/10000.jpg size: 467x499, time: 0.048500776290893555
/data/wt/TestData/dataset/dogsvscats/test1/10001.jpg size: 499x293, time: 0.06814789772033691
/data/wt/TestData/dataset/dogsvscats/test1/10002.jpg size: 500x374, time: 0.06205487251281738
/data/wt/TestData/dataset/dogsvscats/test1/10003.jpg size: 500x263, time: 0.05933856964111328
/data/wt/TestData/dataset/dogsvscats/test1/10004.jpg size: 488x399, time: 0.05417633056640625
/data/wt/TestData/dataset/dogsvscats/test1/10005.jpg size: 249x227, time: 0.04686784744262695
/data/wt/TestData/dataset/dogsvscats/test1/10006.jpg size: 499x375, time: 0.05886483192443848
/data/wt/TestData/dataset/dogsvscats/test1/10007.jpg size: 500x374, time: 0.05744314193725586
/data/wt/TestData/dataset/dogsvscats/test1/10008.jpg size: 499x375, time: 0.05886197090148926
/data/wt/TestData/dataset/dogsvscats/test1/10009.jpg size: 489x367, time: 0.057126522064208984
/data/wt/TestData/dataset/dogsvscats/test1/1001.jpg size: 250x243, time: 0.045355796813964844
/data/wt/TestData/dataset/dogsvscats/test1/10010.jpg size: 500x374, time: 0.058531999588012695
/data/wt/TestData/dataset/dogsvscats/test1/10011.jpg size: 400x345, time: 0.05124783515930176
/data/wt/TestData/dataset/dogsvscats/test1/10012.jpg size: 335x409, time: 0.05216383934020996
/data/wt/TestData/dataset/dogsvscats/test1/10013.jpg size: 470x500, time: 0.04745936393737793
/data/wt/TestData/dataset/dogsvscats/test1/10015.jpg size: 500x387, time: 0.05611777305603027
/data/wt/TestData/dataset/dogsvscats/test1/10016.jpg size: 240x269, time: 0.049646854400634766
/data/wt/TestData/dataset/dogsvscats/test1/10017.jpg size: 250x293, time: 0.05157661437988281
/data/wt/TestData/dataset/dogsvscats/test1/10018.jpg size: 434x375, time: 0.05177617073059082
/data/wt/TestData/dataset/dogsvscats/test1/10019.jpg size: 499x432, time: 0.051323652267456055
/data/wt/TestData/dataset/dogsvscats/test1/1002.jpg size: 374x500, time: 0.05673623085021973
/data/wt/TestData/dataset/dogsvscats/test1/10020.jpg size: 199x200, time: 0.0448911190032959
/data/wt/TestData/dataset/dogsvscats/test1/10021.jpg size: 396x260, time: 0.06324148178100586
/data/wt/TestData/dataset/dogsvscats/test1/10022.jpg size: 500x374, time: 0.057634830474853516
/data/wt/TestData/dataset/dogsvscats/test1/10023.jpg size: 360x269, time: 0.05904746055603027
/data/wt/TestData/dataset/dogsvscats/test1/10024.jpg size: 303x359, time: 0.050252437591552734
/data/wt/TestData/dataset/dogsvscats/test1/10025.jpg size: 443x459, time: 0.04602837562561035
/data/wt/TestData/dataset/dogsvscats/test1/10026.jpg size: 499x375, time: 0.05867409706115723
/data/wt/TestData/dataset/dogsvscats/test1/10027.jpg size: 320x431, time: 0.05827951431274414
/data/wt/TestData/dataset/dogsvscats/test1/10028.jpg size: 200x149, time: 0.05838799476623535
/data/wt/TestData/dataset/dogsvscats/test1/10029.jpg size: 319x414, time: 0.05730867385864258
/data/wt/TestData/dataset/dogsvscats/test1/1003.jpg size: 499x398, time: 0.055377960205078125
/data/wt/TestData/dataset/dogsvscats/test1/10030.jpg size: 479x401, time: 0.05260920524597168
/data/wt/TestData/dataset/dogsvscats/test1/10031.jpg size: 274x500, time: 0.06286954879760742
/data/wt/TestData/dataset/dogsvscats/test1/10033.jpg size: 499x375, time: 0.06049299240112305
/data/wt/TestData/dataset/dogsvscats/test1/10034.jpg size: 406x400, time: 0.04815244674682617
/data/wt/TestData/dataset/dogsvscats/test1/10035.jpg size: 310x458, time: 0.061447858810424805
/data/wt/TestData/dataset/dogsvscats/test1/10036.jpg size: 499x375, time: 0.0570530891418457
/data/wt/TestData/dataset/dogsvscats/test1/10037.jpg size: 499x375, time: 0.06384038925170898
/data/wt/TestData/dataset/dogsvscats/test1/10038.jpg size: 499x466, time: 0.04932117462158203
``

我在i9-9900k和3070上测试得到的延迟是50ms左右. 测试代码：

Copyright (c) OpenMMLab. All rights reserved.

import asyncio from argparse import ArgumentParser

from mmdet.apis import (async_inference_detector, inference_detector, init_detector, show_result_pyplot)

import repvit import torch

from pathlib import Path import time from PIL import Image import magic

def parse_args(): parser = ArgumentParser() parser.add_argument('img', help='Image file') parser.add_argument('config', help='Config file') parser.add_argument('checkpoint', help='Checkpoint file') parser.add_argument('--out-file', default=None, help='Path to output file') parser.add_argument( '--device', default='cuda:0', help='Device used for inference') parser.add_argument( '--palette', default='coco', choices=['coco', 'voc', 'citys', 'random'], help='Color palette used for visualization') parser.add_argument( '--score-thr', type=float, default=0.3, help='bbox score threshold') parser.add_argument( '--async-test', action='store_true', help='whether to set async options for async inference.') args = parser.parse_args() return args

def main(args):

build the model from a config file and a checkpoint file

model = init_detector(args.config, args.checkpoint, device=args.device)

data_dir = Path(args.img)
for fpath in data_dir.iterdir():
# fpaths = [str(fpath) for fpath in data_dir.iterdir()]

    start = time.time()
    # test a single image
    result = inference_detector(model,  str(fpath))
    torch.cuda.synchronize()
    end = time.time()

    # print(f"Inference time: {end-start},  total: {len(fpaths)} average time: {(end-start)/len(fpaths)}")
    img = Image.open(str(fpath))

    print(f"{fpath} size: {img.width}x{img.height}, time: {end-start}")
    img.close()
# show the results
# show_result_pyplot(
#     model,
#     str(fpath),
#     result,
#     palette=args.palette,
#     score_thr=args.score_thr,
#     out_file=args.out_file)

async def async_main(args):

build the model from a config file and a checkpoint file

model = init_detector(args.config, args.checkpoint, device=args.device)
# test a single image
tasks = asyncio.create_task(async_inference_detector(model, args.img))
result = await asyncio.gather(tasks)
# show the results
show_result_pyplot(
    model,
    args.img,
    result[0],
    palette=args.palette,
    score_thr=args.score_thr,
    out_file=args.out_file)

if name == 'main': args = parse_args() if args.async_test: asyncio.run(async_main(args)) else: main(args)

测试结果：

/data/wt/TestData/dataset/dogsvscats/test1/1.jpg size: 381x499, time: 0.08272171020507812 /data/wt/TestData/dataset/dogsvscats/test1/10.jpg size: 500x374, time: 0.06478381156921387 /data/wt/TestData/dataset/dogsvscats/test1/100.jpg size: 499x343, time: 0.07163786888122559 /data/wt/TestData/dataset/dogsvscats/test1/1000.jpg size: 288x499, time: 0.0686182975769043 /data/wt/TestData/dataset/dogsvscats/test1/10000.jpg size: 467x499, time: 0.048500776290893555 /data/wt/TestData/dataset/dogsvscats/test1/10001.jpg size: 499x293, time: 0.06814789772033691 /data/wt/TestData/dataset/dogsvscats/test1/10002.jpg size: 500x374, time: 0.06205487251281738 /data/wt/TestData/dataset/dogsvscats/test1/10003.jpg size: 500x263, time: 0.05933856964111328 /data/wt/TestData/dataset/dogsvscats/test1/10004.jpg size: 488x399, time: 0.05417633056640625 /data/wt/TestData/dataset/dogsvscats/test1/10005.jpg size: 249x227, time: 0.04686784744262695 /data/wt/TestData/dataset/dogsvscats/test1/10006.jpg size: 499x375, time: 0.05886483192443848 /data/wt/TestData/dataset/dogsvscats/test1/10007.jpg size: 500x374, time: 0.05744314193725586 /data/wt/TestData/dataset/dogsvscats/test1/10008.jpg size: 499x375, time: 0.05886197090148926 /data/wt/TestData/dataset/dogsvscats/test1/10009.jpg size: 489x367, time: 0.057126522064208984 /data/wt/TestData/dataset/dogsvscats/test1/1001.jpg size: 250x243, time: 0.045355796813964844 /data/wt/TestData/dataset/dogsvscats/test1/10010.jpg size: 500x374, time: 0.058531999588012695 /data/wt/TestData/dataset/dogsvscats/test1/10011.jpg size: 400x345, time: 0.05124783515930176 /data/wt/TestData/dataset/dogsvscats/test1/10012.jpg size: 335x409, time: 0.05216383934020996 /data/wt/TestData/dataset/dogsvscats/test1/10013.jpg size: 470x500, time: 0.04745936393737793 /data/wt/TestData/dataset/dogsvscats/test1/10015.jpg size: 500x387, time: 0.05611777305603027 /data/wt/TestData/dataset/dogsvscats/test1/10016.jpg size: 240x269, time: 0.049646854400634766 /data/wt/TestData/dataset/dogsvscats/test1/10017.jpg size: 250x293, time: 0.05157661437988281 /data/wt/TestData/dataset/dogsvscats/test1/10018.jpg size: 434x375, time: 0.05177617073059082 /data/wt/TestData/dataset/dogsvscats/test1/10019.jpg size: 499x432, time: 0.051323652267456055 /data/wt/TestData/dataset/dogsvscats/test1/1002.jpg size: 374x500, time: 0.05673623085021973 /data/wt/TestData/dataset/dogsvscats/test1/10020.jpg size: 199x200, time: 0.0448911190032959 /data/wt/TestData/dataset/dogsvscats/test1/10021.jpg size: 396x260, time: 0.06324148178100586 /data/wt/TestData/dataset/dogsvscats/test1/10022.jpg size: 500x374, time: 0.057634830474853516 /data/wt/TestData/dataset/dogsvscats/test1/10023.jpg size: 360x269, time: 0.05904746055603027 /data/wt/TestData/dataset/dogsvscats/test1/10024.jpg size: 303x359, time: 0.050252437591552734 /data/wt/TestData/dataset/dogsvscats/test1/10025.jpg size: 443x459, time: 0.04602837562561035 /data/wt/TestData/dataset/dogsvscats/test1/10026.jpg size: 499x375, time: 0.05867409706115723 /data/wt/TestData/dataset/dogsvscats/test1/10027.jpg size: 320x431, time: 0.05827951431274414 /data/wt/TestData/dataset/dogsvscats/test1/10028.jpg size: 200x149, time: 0.05838799476623535 /data/wt/TestData/dataset/dogsvscats/test1/10029.jpg size: 319x414, time: 0.05730867385864258 /data/wt/TestData/dataset/dogsvscats/test1/1003.jpg size: 499x398, time: 0.055377960205078125 /data/wt/TestData/dataset/dogsvscats/test1/10030.jpg size: 479x401, time: 0.05260920524597168 /data/wt/TestData/dataset/dogsvscats/test1/10031.jpg size: 274x500, time: 0.06286954879760742 /data/wt/TestData/dataset/dogsvscats/test1/10033.jpg size: 499x375, time: 0.06049299240112305 /data/wt/TestData/dataset/dogsvscats/test1/10034.jpg size: 406x400, time: 0.04815244674682617 /data/wt/TestData/dataset/dogsvscats/test1/10035.jpg size: 310x458, time: 0.061447858810424805 /data/wt/TestData/dataset/dogsvscats/test1/10036.jpg size: 499x375, time: 0.0570530891418457 /data/wt/TestData/dataset/dogsvscats/test1/10037.jpg size: 499x375, time: 0.06384038925170898 /data/wt/TestData/dataset/dogsvscats/test1/10038.jpg size: 499x466, time: 0.04932117462158203

请问一下，这个作为backbone放到检测模型中，是不是比resnet系列会快一些