search

YiwuZhong / Sub-GC

[ECCV 2020] Official code for "Comprehensive Image Captioning via Scene Graph Decomposition"
MIT License
96 stars 16 forks source link

how to use myself dataset？ #7

Closed zxh19976666 closed 3 years ago

zxh19976666 commented 3 years ago

thanks your great work sir！ but I want to use myself dataset？ could you help me sir？ plz ！thanks！

YiwuZhong commented 3 years ago

Since this work was trained and evaluated on the standard image captioning benchmark (e.g., COCO Caption and Flickr30k), we don't plan to provide code for custom dataset.

PS: If you want to build a captioning model on your own dataset, you will need to process the captions into the labels in the same format as COCO/Flickr30k datasets. Also, you have to obtain the region features from the object detector, the scene graphs from Motif-Net, and the sampled sub-graphs for all the images in your dataset.

zxh19976666 commented 3 years ago

thank you for your reply sir！