I can't achieve your mAP (77.x)

[kimna4]

I tried to run many times with different parameters. But I could not achieve the reported mAP (over 77%).

Almost mAP which I get are 71.x%.

Could you tell me how to set the training environment??

Thank you in advance.

[sunsiyang2015]

I tried to run many times with different parameters. But I could not achieve the reported mAP (over 77%).

Almost mAP which I get are 71.x%.

Could you tell me how to set the training environment??

Thank you in advance. I also can not achieve mAP 77%, which is only 71%. Have you solve the problem?

[SchafferZhang]

same!

[kimna4]

@sunsiyang2015 @SchafferZhang I am using this SSD ( https://github.com/qijiezhao/pytorch-ssd ) and can achieve over 77.2% with many tries.

However I think that the SSD of qijiezhao is also one of the branch version of amdegroot's SSD. So... if we know parameters such as iteration#, leraning rate. We can get high mAP with amdegroot's SSD.

[niaoyu]

77.29 for mine

[SchafferZhang]

@niayou Do you modify any parameter ?

[ZhipengLiu6]

I achieve map 77.44 in pytorch0.3 and 69.34 in pytorch1.0 with default parameters.

[door5719]

Mean AP = 0.7743

Results:
0.817
0.854
0.757
0.696
0.499
0.848
0.857
0.874
0.614
0.823
0.792
0.856
0.871
0.847
0.789
0.500
0.774
0.791
0.862
0.763
0.774

[door5719]

VOC2017 DATESET TEST

[lesonly]

77.29 for mine I can not achieve mAP 77%, which is only 71%. Do you modify any parameter ?

[chenaifang]

Which model did you test for your results? ssd300_map_77.43_v2.pth published by the author or is it a model that you've trained on the basis of VGG weights @ZhipengLiu6 @SchafferZhang @niaoyu @sunsiyang2015 @kimna4 @all the person

[Bobby2090]

ng this SSD ( https://github.com/qijiezhao/pytorch-ssd ) and can achieve over 77.2% with many tries.

However

你训练SSD最终loss大概是多少呢？谢谢

[ksmdnl]

did you train your model from scratch or upload the pre trained weights on your model? maybe that's one thing to consider?