Fine tuning a pretrained tensorflow model to generate image captions -
i fine tuning pretrained im2txt model in tensorflow explained in tensorflow models-im2txt tensorflow models-im2txt.
i use new image+sentence dataset while pretrained model(2m steps) trained on ms coco dataset. 1m fine-tuning steps didn't end yet tested latest checkpoints images. unfortunately, captions ". ", "and " , etc. word counts, used new wordcounts new dataset guess mistake. new deep learning , tensorflow, explain, how should use wordcounts? should use 1 ms coco, or fine-tuning dataset or combination? or missing whole point or crucial?
Comments
Post a Comment