pytorch 序列化性能测试

版权声明：本文为博主原创文章，未经博主允许不得转载。 https://blog.csdn.net/jacke121/article/details/84347970

如果单张图片比，torch比opencv读写要快。但是数据量大了之后，也慢，100张图片一起，600ms左右。

import time
from distributed.protocol import serialize, deserialize
import cv2
import torch
import torch.nn.functional as f
if __name__ == '__main__':
    obj={'mat':torch.randn(10, 10),'name': '10','test':{'entry':1}}
    torch.save(obj,'test.dat' )
    for i in range(1000):
        start = time.time()
        x = torch.rand(100, 3, 352, 352)
        print('time1', time.time() - start)  # 需要8ms左右。
        ser1=serialize(x)
        print('time2', time.time() - start)  # 需要8ms左右。
        t2 = deserialize(*ser1)
        # torch.save(tensor,'d:/img/test'+str(i)+'.dat')#io需要10ms，opncv需要24ms左右
        print('time3',time.time()-start)#需要8ms左右。
        start = time.time()
        for i in range(100):
            x = torch.rand(1, 3, 352, 352)
            ser1 = serialize(x)
            t2 = deserialize(*ser1)
        print('time4', time.time() - start)  # 需要8ms左右。

pytorch的for循环还是做了一点优化的：

time1 0.2355334758758545
time2 0.45803236961364746
time3 0.6065003871917725
time4 0.6360311508178711
time1 0.2174680233001709
time2 0.4389991760253906
time3 0.5855002403259277
time4 0.623499870300293
time1 0.21796584129333496
time2 0.4394986629486084
time3 0.5860018730163574
time4 0.6004657745361328
time1 0.2174983024597168
time2 0.4370310306549072
time3 0.5840315818786621
time4 0.5950057506561279
time1 0.22396206855773926
time2 0.44996213912963867
time3 0.5954670906066895
time4 0.6134955883026123
time1 0.2265324592590332
time2 0.45000123977661133
time3 0.595501184463501
time4 0.6365303993225098
time1 0.21999859809875488
time2 0.4399685859680176
time3 0.5854685306549072
time4 0.6674997806549072
time1 0.2175295352935791
time2 0.4419982433319092
time3 0.5939993858337402
time4 0.720999002456665
time1 0.26000189781188965
time2 0.4855027198791504
time3 0.634000301361084
time4 0.616499662399292
time1 0.2240004539489746
time2 0.4440333843231201
time3 0.5910031795501709
time4 0.6299972534179688
time1 0.21850204467773438
time2 0.43353867530822754
time3 0.5790259838104248
time4 0.6274731159210205
time1 0.2174997329711914
time2 0.44050002098083496
time3 0.5919997692108154
time4 0.6039998531341553
time1 0.21899962425231934
time2 0.44749975204467773
time3 0.5920000076293945
time4 0.6035275459289551
time1 0.22899961471557617
time2 0.44597458839416504
time3 0.5970087051391602
time4 0.6219675540924072
time1 0.21902751922607422
time2 0.43403077125549316
time3 0.5775296688079834
time4 0.6179683208465576
time1 0.2615315914154053
time2 0.4770326614379883
time3 0.6215341091156006
time4 0.6079990863800049
time1 0.2174668312072754
time2 0.4395122528076172
time3 0.5859675407409668
time4 0.6054995059967041
time1 0.2225019931793213
time2 0.44553184509277344
time3 0.5910322666168213
time4 0.610999345779419
time1 0.23100042343139648
time2 0.4459669589996338
time3 0.5890064239501953
time4 0.639991283416748
time1 0.2190086841583252
time2 0.43796873092651367
time3 0.5825061798095703
time4 0.6484625339508057
time1 0.21753406524658203
time2 0.43250203132629395
time3 0.5775046348571777
time4 0.7499957084655762
time1 0.22949957847595215
time2 0.4584989547729492
time3 0.6049997806549072
time4 0.620999813079834
time1 0.23253846168518066
time2 0.4545013904571533
time3 0.6105022430419922
time4 0.6355316638946533
time1 0.21899962425231934
time2 0.4570002555847168
time3 0.6060004234313965
time4 0.6459980010986328
time1 0.21749329566955566
time2 0.43249940872192383
time3 0.57846999168396
time4 0.600039005279541
time1 0.21645879745483398
time2 0.4394817352294922
time3 0.5874612331390381
time4 0.6480300426483154
time1 0.22749876976013184
time2 0.4464995861053467
time3 0.5905001163482666
time4 0.6440064907073975
time1 0.22399067878723145
time2 0.4465053081512451
time3 0.5919601917266846