Training iterations of FixMatch on the Semi-Aves

[LiheYoung]

Hi,

You mentioned in your paper that you train FixMatch for 500 epochs on the Semi-Aves dataset. Since the labeled batch size is 32 and there are totally 3,959 labeled images, the training iterations are around 50,000. Is it right? However, when I train the FixMatch for 50,000 iterations from scratch, the performance is only around 6%, much lower that the 28% you reported. Is there anything I missed? Thanks.