MCG-NJU / MultiSports

[ICCV 2021] MultiSports: A Multi-Person Video Dataset of Spatio-Temporally Localized Sports Actions
Other
109 stars 7 forks source link

Anchors used for YoWo Benchmark #16

Open Griffintaur opened 2 years ago

Griffintaur commented 2 years ago

I am trying to reproduce the result using YOWO? Can you let me know which anchors you used for YOWO benchmarks?

MiaSanLei commented 2 years ago

I am trying to reproduce the result using YOWO? Can you let me know which anchors you used for YOWO benchmarks?

The anchors we used are [0.70458, 1.18803, 1.26654, 2.55121, 1.59382, 4.08321, 2.30548, 4.94180, 3.52332, 5.91979] as same as the anchors of ucf24 dataset in YOWO's codebase.

Griffintaur commented 2 years ago

Which pretrained model you used to initialize the 3D and 2D model in YoWo architecture?

MiaSanLei commented 2 years ago

Which pretrained model you used to initialize the 3D and 2D model in YoWo architecture?

We use Darknet-19 weights as our 2D pretrained model and resnext-101-kinetics.pth from Google drive as 3D pretrained.

Griffintaur commented 2 years ago

@MiaSanLei sorry to ask another question, I was using the above settings as you described but nan was creeping in during training. Did you change anything in ucf.yaml config file or anything to make it work for Multisports. It would be wonderful if you could share the config file you used.

MiaSanLei commented 2 years ago

I modified the learning rate and classes in ucf24.cfg.

------------------ 原始邮件 ------------------ 发件人: "MCG-NJU/MultiSports" @.>; 发送时间: 2022年2月16日(星期三) 上午10:24 @.>; @.**@.>; 主题: Re: [MCG-NJU/MultiSports] Anchors used for YoWo Benchmark (Issue #16)

@MiaSanLei sorry to ask another question, I was using the above settings as you described but nan was creeping in during training. Did you change anything in ucf.yaml config file or anything to make it work for Multisports. It would be wonderful if you could share the config file you used.

— Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android. You are receiving this because you were mentioned.Message ID: @.***>