research model 'ssd_resnet_50_fpn_coco' with input shape [?,?,?,3]

Hello everyone.

I am using tensorflow/models for model translation and the model is 'ssd_resnet_50_fpn_coco'. In the translation process, i want to know the exact value of the input shape, such as [1,640,640,3], but the input shape of 'ssd_resnet_50_fpn_coco' is [?,?,?,3]. So i want to know how to generate a new 'ssd_resnet_50_fpn_coco' model with input shape [1,640,640,3]?

I have tried to use pipeline.config of 'ssd_resnet_50_fpn_coco' to train again, but the result model was not the same as your research model 'ssd_resnet_50_fpn_coco(In fact, the featureextractor modules were the same, but other modules were not the same).

And in the file pipeline.config, i only modified PATH_TO_BE_CONFIGED.