COCO2014 or COCO2017 ?

[Jarvis73]

Hi Bob,

Good work on few-shot segmentation. But I have a question about the COCO dataset.

I think PANet and FWB are both evaluated on COCO 2014(~83K training images), but it seems that PMMs are evaluated on COCO 2017(~118K training images). See the code coco_train.py.

If true, I think the comparison in Table 6 (in the paper) is not fair. If I missed some details, please let me know.

Thank you .

[Yang-Bob]

Hi,

We have conducted experiments on both COCO 2014 and COCO 2017 and we find that the performances are similar.
Based on our result, I don't think the version of dataset has a significant impact.

Best regards

[Jarvis73]

@Yang-Bob Thank you for your reply. The similar performances caused by either the underfitting or the few-shot task itself may deserve to be discussed in the future.