Invalid loss, terminating training


This is my train_kitty.txt.



The size of my input image is 1920, 1080 and I resize it to 1920, 1056.The size of my input image is 1920, 1080 and I resize it to 1920, 1056. I have successfully run faster-rcnn, and retinanet is currently no problem.

Please try a smaller max_learning_rate.
For example, 1e-2

Will different models be trained at the same time using the kitty format data converted through tlt-dataset-convert under the same path will interfere with each other (for example, yolov3 and retinanet use DATA_DOWNLOAD_DIR/tfrecords/kitti_trainval/kitti_trainval at the same time)

The training dataset can be used for both yolov3 and retinanet.