how to predict end to end

[AutumnZ-94]

I notice the code in predict.py that cids_of_imgs imgs.embeddedG.npy imgs.embedded.npy sids_of_txts txts.embeddedG.npy txts.embedded.npy are required for predict and evaluation.

graph.get_tensor_by_name("inputs_seq:0"): inputs_seq_batch, graph.get_tensor_by_name("inputs_seq_len:0"): inputs_seq_len_batch, graph.get_tensor_by_name("inputs_seq_embedded:0"): inputs_seq_embedded_batch, graph.get_tensor_by_name("inputs_seq_embeddedG:0"): inputs_seq_embeddedG_batch, graph.get_tensor_by_name("inputs_img_embedded:0"): inputs_img_embedded_batch, graph.get_tensor_by_name("inputs_img_embeddedG:0"): inputs_img_embeddedG_batch, graph.get_tensor_by_name("dropout_prob:0"): 0

So how could I use it in real world to predict ?
Whether all these input tensor are necessary without evaluation， should I switch the model to an inference model?

[wavewangyue]

Yes, all these input tensor are necessary in prediction (or evaluation)