search

sfzhang15 / ATSS

Bridging the Gap Between Anchor-based and Anchor-free Detection via Adaptive Training Sample Selection, CVPR, Oral, 2020
Other
1.08k stars 162 forks source link

what is your final model used by ATSS in table 8? #60

Closed HyogiJeon closed 4 years ago

HyogiJeon commented 4 years ago

Hello.

I saw your paper and I have 1 question. What is your final model used by ATSS in table 8?

And I also saw the README. You posted (i.e., consistent with models in FCOS and maskrcnn_benchmark). Which model do you used for ATSS? FCOS? or maskrcnn?

Thank you in advances for your help!

sfzhang15 commented 4 years ago

@HyogiJeon

  1. Our final models in Table 8 are the improved RetinaNet (#A=1) with ATSS.
  2. ATSS uses the codes of FCOS to implement, and FCOS uses the codes of maskrcnn to implement.
HyogiJeon commented 4 years ago

Thank you for your answer. And I have some additional question.

  1. You said that the final model is the improved RetinaNet (#A=1) with ATSS. Then, is it right the same model in Table 3 and in Table 8? But, they have difference MAP.(39.3 in Table 3, 43.6 in Table 8)

  2. Is it right the model posted in a paper and the model in your code are difference?

Thank you.

sfzhang15 commented 4 years ago

@HyogiJeon

  1. Yes. Please see the first paragraph in Sec 4.4 for their training differences.
  2. No, it is the same.