A little bug of the function text2image_recall_at_k in evaluator.py

[XuanxuanGao]

Hi, I'm working on text-image retrieval and I appreciate your tensorflow version of vesepp model.

During the code review, I'm confused about the implementation of the function text2image_recall_at_k in evaluator.py. The original code is shown as below:

vsepp_tensorflow/src/evaluators.py

Line 117 in 30c8193

for index in range(num_captions):

While I think that line 117 should be like this, since the embedded_captions cannot be sliced 5 fold:

for index in range(num_captions//5):

Interestingly, this change does not change the final evaluation result.

[gorjanradevski]

Hey @XuanxuanGao, thanks for the interest. I will have a look these days and let you know if indeed there is a bug. In the meantime, have a look at: https://github.com/fartashf/vsepp/blob/master/evaluation.py#L258

This is the original implementation, and the method you are looking at is just a modified version.