DavidHuji / CapDec

CapDec: SOTA Zero Shot Image Captioning Using CLIP and GPT2, EMNLP 2022 (findings)

DavidHuji/CapDec Issues

Why noise is added to “prefix”（visual embeding by clip） instead of token（text embeding by tokenizer）in the code？
Closed 7 months ago1
Error(s) in loading state_dict for ClipCaptionModel
Updated 9 months ago
modality_offset
Updated 9 months ago1
how to download / create single_caption_per_sample_val.json file
Updated 9 months ago6
how to get train.json?
Closed 9 months ago1
About FlickrStyle10K
Updated a year ago1
About using the Sentcap dataset and setting parameters for train.py
Updated a year ago5
training parameters on Flickr30k dataset
Closed a year ago2
about the performance of the Zerocap demonstrated in this paper
Updated a year ago1
about the parameters of train.py
Updated a year ago1
Different results at CiDER metric in paper
Updated a year ago1
About the type of gpt2
Updated a year ago1
annotations files of coco during evaluation
Closed 2 years ago1
About the variance of 0.016
Updated 2 years ago3
Requirements File
Updated 2 years ago3