When i train the model on the mini or trainval dataset using the simple detr3d_resnet101_gridmask config , i get around 0 map and 0 nds in the end, the model does not learn anything as the training process goes, but evaluating the model using your trained weights i can get the correct results, I am using nuscenes mini v1.0 dataset with mmdetection 1.0.0rc2 mmcv 1.6.0 and pytorch 1.9.1 with cuda 11.1
When i train the model on the mini or trainval dataset using the simple detr3d_resnet101_gridmask config , i get around 0 map and 0 nds in the end, the model does not learn anything as the training process goes, but evaluating the model using your trained weights i can get the correct results, I am using nuscenes mini v1.0 dataset with mmdetection 1.0.0rc2 mmcv 1.6.0 and pytorch 1.9.1 with cuda 11.1