ssd-shufflenet mAP ?

[ananddb90]

Thank you for your work. Do you have training results of SSD detection on shuffle-net because I would like to know mAP thank you

[linchaozhang]

@ananddb90 not good, only 63% on VOC, still in finetune. how about you?

[ananddb90]

I tried with RFCN and faster-RCNN and I got very less on VOC (~43%) and story was when I used mobilenet. where as ssd on mobile net performed really good (~72) . have faced similar issues before ? Also could you please tell me how you are planning to fine tune further ? may be I am missing something regards, anand

[zoufangyu1987]

@linchaozhang @ananddb90 Do you have 0.5_ShuffleNet or 0.75_ShuffleNet pre_trained caffe model ? and its train.prototxt? Thank you very much!

[ananddb90]

@zoufangyu1987 sorry I don't have 0.5_ShuffleNet or 0.75_ShuffleNet models

[CrazySssst]

@linchaozhang Do you pre-train the model in coco?

[linchaozhang]

@CrazySssst i have pre-trained on coco, 63% is the result after pre-train on coco.

[ghost]

Hi, is there any chance you can provide the pretrained models?

[share020]

@ananddb90 @linchaozhang Hello, have you succeed in training the shufflenet-ssd with this code? I tried it with pascal voc dataset and the model didn't converge after 120000 iteration. Could you please tell me what's wrong and provide your pretrained model? Thanks.

[zfc929]

I train my data ,the map is 0.658, but the small object is not well ,who train model with 512 *512 ?

[gjm441]

I trained shufflenet-ssd on widerface, easy and medium AP only got 0.6 .It got bad performance, the trainning loss stayed at 3.7 at last and hard to descend .May be shufflenet not suitable for detection? @linchaozhang give some suggestions?