I remember that some time ago, the "Institute of computer vision" pushed a document on the best classification of CVPR2020 (link: CVPR2020 best target detection | AdderNet (additive network) includes papers and source code links ), one of the students asked, can this new classification framework be grafted into the detection network, will it be improved? Today, we will tell you how to improve through experiments?

Prospect review

It is estimated that some students have forgotten the framework and essence of the addition network. Let's simply return to the specific framework details first.

Researchers and developers are used to taking convolution as the default operation, extracting features from visual data, and introducing various methods to accelerate convolution, even if there is a risk of sacrificing network capability. But few people try to replace convolution with another more effective similarity measure. In fact, the computational complexity of addition is much lower than that of multiplication. Therefore, the author is motivated to study the feasibility of using addition in convolutional neural network instead of multiplication.

Different convolution visualization results can be reviewed below:

The specific code flow is as follows:

Next, we will start to modify its code and graft it into the yolov3 framework to see what wonderful changes can happen?

Pytorch Yolo should be installed and can be trained and tested. Now let's start directly! First of all, let me take part of the code of pytorch-yolov3 as an example:

In the above, the Darknet Backbone network is directly written in the detection part. In fact, such a framework is not good and tends to be customized, so you can write a Backbone function and DetectHead function by yourself. It makes sense to write in this way. You can see why you want to write in this way?

In this way, the backbone is a module. You can call the backbone network you want at will, and you can also extract the corresponding detection header at will. Next, let's look at the code of AdderNet:

According to the paper (link: AdderNet link )Modify the corresponding convolution, and then apply it to ResNet50 network this time. Let's replace Conv in ResNet50. Let's continue:

The next step is a simple step. Replace the modified Backbone network, and then train and test in the corresponding data set to observe the differences between them. In order to observe more carefully, the specific output results are not given in detail, as follows:

+ Class '0' (person) - AP: 0.69071601970752
+ Class '1' (bicycle) - AP: 0.4686961863448047
+ Class '2' (car) - AP: 0.584785409652401
+ Class '3' (motorbike) - AP: 0.6173425471546101
+ Class '4' (aeroplane) - AP: 0.7368216071089109
+ Class '5' (bus) - AP: 0.7522709365644746
+ Class '6' (train) - AP: 0.754366135549987
+ Class '7' (truck) - AP: 0.4188454158138422
+ Class '8' (boat) - AP: 0.4055367699507446
+ Class '9' (traffic light) - AP: 0.44435250125992093
+ Class '10' (fire hydrant) - AP: 0.7803236133317674
+ Class '11' (stop sign) - AP: 0.7203250980406222
+ Class '12' (parking meter) - AP: 0.5318708513711929
+ Class '13' (bench) - AP: 0.33347708090637457
+ Class '14' (bird) - AP: 0.4441360921558241
+ Class '15' (cat) - AP: 0.7303504067363646
+ Class '16' (dog) - AP: 0.7319887348116905
+ Class '17' (horse) - AP: 0.77512155236337
+ Class '18' (sheep) - AP: 0.5984679238272702
+ Class '19' (cow) - AP: 0.5233874581223704
+ Class '20' (elephant) - AP: 0.8563788399614207
+ Class '21' (bear) - AP: 0.7462024921293304
+ Class '22' (zebra) - AP: 0.7870769691158629
+ Class '23' (giraffe) - AP: 0.8227873134751092
+ Class '24' (backpack) - AP: 0.32451636624665287
+ Class '25' (umbrella) - AP: 0.5271238663832635
+ Class '26' (handbag) - AP: 0.20446396737325406
+ Class '27' (tie) - AP: 0.49596217809096577
+ Class '28' (suitcase) - AP: 0.569835653931444
+ Class '29' (frisbee) - AP: 0.6356266022474135
+ Class '30' (skis) - AP: 0.40624013441992135
+ Class '31' (snowboard) - AP: 0.4548600158139028
+ Class '32' (sports ball) - AP: 0.5431383703116072
+ Class '33' (kite) - AP: 0.4099711653381243
+ Class '34' (baseball bat) - AP: 0.5038339063455582
+ Class '35' (baseball glove) - AP: 0.47781969136825725
+ Class '36' (skateboard) - AP: 0.6849120730914782
+ Class '37' (surfboard) - AP: 0.6221252845246673
+ Class '38' (tennis racket) - AP: 0.68764570668767
+ Class '39' (bottle) - AP: 0.4228582945038891
+ Class '40' (wine glass) - AP: 0.5107649160534952
+ Class '41' (cup) - AP: 0.4708999794256628
+ Class '42' (fork) - AP: 0.44107168135464947
+ Class '43' (knife) - AP: 0.288951366082318
+ Class '44' (spoon) - AP: 0.21264460558898557
+ Class '45' (bowl) - AP: 0.4882936721018784
+ Class '46' (banana) - AP: 0.27481021398716976
+ Class '47' (apple) - AP: 0.17694573390321539
+ Class '48' (sandwich) - AP: 0.4595098054471395
+ Class '49' (orange) - AP: 0.2861568847973789
+ Class '50' (broccoli) - AP: 0.34978362407336433
+ Class '51' (carrot) - AP: 0.22371776472064184
+ Class '52' (hot dog) - AP: 0.3702692586995472
+ Class '53' (pizza) - AP: 0.5297757751733385
+ Class '54' (donut) - AP: 0.5068384767127795
+ Class '55' (cake) - AP: 0.476632708387989
+ Class '56' (chair) - AP: 0.3980449296511249
+ Class '57' (sofa) - AP: 0.5214086539073353
+ Class '58' (pottedplant) - AP: 0.4239751120301045
+ Class '59' (bed) - AP: 0.6338351737747959
+ Class '60' (diningtable) - AP: 0.4138012499478281
+ Class '61' (toilet) - AP: 0.7377284037968452
+ Class '62' (tvmonitor) - AP: 0.6991588571748895
+ Class '63' (laptop) - AP: 0.68712851664284
+ Class '64' (mouse) - AP: 0.7214480416511962
+ Class '65' (remote) - AP: 0.4789729416954784
+ Class '66' (keyboard) - AP: 0.6644829934265277
+ Class '67' (cell phone) - AP: 0.39743578548434444
+ Class '68' (microwave) - AP: 0.6423763095621656
+ Class '69' (oven) - AP: 0.48313299304876195
+ Class '70' (toaster) - AP: 0.16233766233766234
+ Class '71' (sink) - AP: 0.5075074098080213
+ Class '72' (refrigerator) - AP: 0.6862896780296917
+ Class '73' (book) - AP: 0.17111744621852634
+ Class '74' (clock) - AP: 0.6886459682881512
+ Class '75' (vase) - AP: 0.44157962279267704
+ Class '76' (scissors) - AP: 0.3437987832196098
+ Class '77' (teddy bear) - AP: 0.5859590979304399
+ Class '78' (hair drier) - AP: 0.11363636363636365
+ Class '79' (toothbrush) - AP: 0.2643722437438991

The visualization results are as follows (there are still great differences):

© The Ending