Open cc19860606 opened 3 years ago
Anyone reproduct the performance of CoordAttention with MobileNetV2 on VOC2007?
Have you pretrained the model on ImageNet?
I firstly test MobileNetV2,the mAP can achieve 71.7. Then, I try to add CoordAttention as shown above, meanwhile the pretrained model of 'mobilenet_v2-b0353104.pth' is not use. The mAP only can achieve 64. @Andrew-Qibin
ssdlite-pytorch-mobilenext with CoordAttention need to be trained with ImageNet, then test on VOC?
Exactly. Otherwise, it does not work. Recently, I will release the pretrained models.
I look forward for it, thx.
The ImageNet dataset used for training is ILSVRC2012 or others? And the hyperparameter and code of data augmentation for training? @Andrew-Qibin
I modify the code of MobileNetV2 by adding CoordAttention at line 101, is it right? @Andrew-Qibin