Open b03505036 opened 3 years ago
@b03505036 Yes. We may release several new models soon.
@b03505036 Yes. We may release several new models soon.
Thank you for reply! But why the SOLOv2 3x in the paper is 38.8 , but us 37.5 in the website, is the gap causing by test-dev?
@b03505036 The model tested on test-dev is trained with 6x schedule. And yes, for our models the test-dev result is usually 0.5~0.8 AP better than val.
@b03505036 The model tested on test-dev is trained with 6x schedule. And yes, for our models the test-dev result is usually 0.5~0.8 AP better than val.
Hello, may I ask what you mean about 6x schedule?
@b03505036 The model tested on test-dev is trained with 6x schedule. And yes, for our models the test-dev result is usually 0.5~0.8 AP better than val.
Hello, may I ask what you mean about 6x schedule?
The model presented in NIPS2020 is 6x.
Could sem used in CondInst apply on SOLO V2 theoretically?