IJB-B and C Evaluation Problem

I am trying to evaluate the performance of my model on IJB-B dataset. I am using the code in the link below:
https://github.com/deepinsight/insightface/blob/master/evaluation/IJB/IJB_11.py

However, i can see that in get_image_feature, there is no call to forward_db function, so it gives me an output with the shape (227630,2) which is wrong. It should be (227630,512) as 512 is the embedding dimension. Then, I tried the following code:
https://github.com/deepinsight/insightface/blob/master/recognition/partial_fc/pytorch/IJB/IJB_11_Batch.py
It generates the embedding with correct dimension, but the evaluation results are very poor.

Methods | 1e-06 | 1e-05 | 0.0001 | 0.001 | 0.01 | 0.1 |
+------------------+-------+-------+--------+-------+------+-------+
| insightface-IJBB | 0.00 | 0.00 | 0.05 | 0.30 | 2.19 | 19.59 |
+------------------+-------+-------+--------+-------+------+-------+

Can someone tell me how to solve this problem? Thank you..

Start with evaluation/IJB/IJB_11.py, especially get_image_feature and the missing forward_db call, then compare it with recognition/partial_fc/pytorch/IJB/IJB_11_Batch.py. Reproduce the reported embedding shapes and evaluation metrics; done means the embeddings have shape (227630,512) and the IJB-B evaluation produces valid results rather than the reported poor scores.

Written by the indexing model from the issue text.