javi-clear-image-ai commented on issue #5228: Good training and validation accuracy, poor testing accuracy URL: https://github.com/apache/incubator-mxnet/issues/5228#issuecomment-436560631 I have the same issue... **TRAIN-VALIDATION:** json_stats: {**"accuracy_cls": 0.977055**, "eta": "0:00:00", "iter": 89999, **"loss": 0.097330**, "loss_bbox": 0.015025, "loss_cls": 0.075462, "loss_rpn_bbox_fpn2": 0.000000, "loss_rpn_bbox_fpn3": 0.000000, "loss_rpn_bbox_fpn4": 0.000153, "loss_rpn_bbox_fpn5": 0.000106, "loss_rpn_bbox_fpn6": 0.000000, "loss_rpn_cls_fpn2": 0.000099, "loss_rpn_cls_fpn3": 0.000064, "loss_rpn_cls_fpn4": 0.000068, "loss_rpn_cls_fpn5": 0.000033, "loss_rpn_cls_fpn6": 0.000000, "lr": 0.000020, "mb_qsize": 64, "mem": 8932, "time": 0.372048} **TESTING:** INFO task_evaluation.py: 186: copypaste: AP,AP50,AP75,APs,APm,APl INFO task_evaluation.py: 187: copypaste: -1.0000,-1.0000,-1.0000,-1.0000,-1.0000,-1.0000 And when I visualize the testing image results, only 4 out of 1000 show up, and some of the classes go wrong ... I am gonna try to change the visualization threshold from 0.9 to 0.4, to see what happens.... Any help would be very much appreciated
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
