Help Training RetinaNet

Hello folks, I’m looking for some help in training a custom RetinaNet model with a Resnet-101 backbone.
I have my dataset in coco format, I’ve tried a few implementations such as keras and others on GitHub but I’m either not seeing the model learn, and the model training validation losses are not being saved.
If anyone knows where I can look to follow an implementation that works, if you could let me know that will be great and thanks in advance. I want to save the model after each epoch, I want to track the mAP, I want to save the training and validation losses so I can plot them. Thanks folks.