MMFewshot evaluate only one class out of two Novel classes

[imadgohar]

I am using TFA, the dataset is in VOC format. I have total 5 classes and in three splits I am dividing it like this ALL CLASSES=[A,B,C,D,E]

Split_1_base=[A,B,C,D] Split_2_base=[A,B,C] Split_3_base=[A,B]

Split_1_novel=[E] Split_2_novel=[D,E] Split_3_novel=[C,D,E]

Class E has few samples that's why it is considered as novel class in split_1. The second class with least samples is D and it is together with E in split 2 and the third class with least samples is C and it is combined with D and E in split_3.

The issue now is in split_2 and split_3, the evaluating results for C and D are always 0. I have the info of GT and Dets but recall and ap is 0.0.

Note: Alphabets are used for classes for simplicity.