When I move the project to Jetson-TX1&2, the mAP is much lower than the situation in my laptop. I trained the model on my own dataset. There is cuda8 and cudnn6 in my TX1, and there is cuda9 and cudnn7 in my TX2.
I followed the suggestion to change the crop_pooling_layer to roi_pooling_layer, and the result mAP is a little higher than before(before is 0, now is 24%), but it is still much lower than the mAP in laptop(62%).
Did you meet this problem ?
When I move the project to Jetson-TX1&2, the mAP is much lower than the situation in my laptop. I trained the model on my own dataset. There is cuda8 and cudnn6 in my TX1, and there is cuda9 and cudnn7 in my TX2. I followed the suggestion to change the crop_pooling_layer to roi_pooling_layer, and the result mAP is a little higher than before(before is 0, now is 24%), but it is still much lower than the mAP in laptop(62%). Did you meet this problem ?