search

WangJiuniu / DistinctiveCap

6 stars 1 forks source link

Could please release the code of finding the similar image sets? The CIDEr one and the proposed one in the paper. #2

Open mad-red opened 3 years ago

mad-red commented 3 years ago

The CIDEr one and the proposed one in the paper.

WangJiuniu commented 3 years ago

The code needs to be cleaned. And this part (i.e., image to text retrieval) is implemented based on https://github.com/ruotianluo/DiscCaptioning . The DiscCaptioning contains the image to text retrieval.

In addition, the CIDEr one is implemented based on the CIDEr metric evaluation code. We run the code on cpu cluster since the large computational complex.

The code will be released in several months I think.

Thank you for your concern.

mad-red commented 3 years ago

The code needs to be cleaned. And this part (i.e., image to text retrieval) is implemented based on https://github.com/ruotianluo/DiscCaptioning . The DiscCaptioning contains the image to text retrieval.

In addition, the CIDEr one is implemented based on the CIDEr metric evaluation code. We run the code on cpu cluster since the large computational complex.

The code will be released in several months I think.

Thank you for your concern.

To find the similar image of image A, do you use 1 sentence of image B as candidate and all sentences of image A as references, and average the computed CIDEr scores over 5 sentences of B to get the final CIDEr similarity of image A and B？And are the CIDEr scores of the top-5 similar images you provided from small to large order？ Sorry to bother again.

WangJiuniu commented 3 years ago

For CIDEr similarity, we use 5 sentences of image A and 5 sentences of image B, and average the 25 values.

Top-5 similar images are not in order, while top-500 similar images are in order.

Thank you!