yjh0410
commented
3 years ago Wow！Wonderful！
Your jobs is very significant.
I must admit that I do have a misunderstanding of DLA-34. I will try this efficient network in the future.
Thanks a lot.
------------------ 原始邮件 ------------------ 发件人: "yjh0410/CenterNet-plus" @.>; 发送时间: 2021年6月30日(星期三) 晚上10:47 @.>; 抄送: "ら @.**@.>; 主题: [yjh0410/CenterNet-plus] About DLA-34 (#1)
@yjh0410 Hi!
You did great job.
I did the benchmark test to find optimal backbone in terms of the execution time and imagenet top1 accuracy.
https://github.com/developer0hye/pytorch-backbone-benchmark
You can see DLA-34 is not heavy in comparison to other networks.
But you mentioned
CenterNet is an encoder-decoder network, but I won't consider Hourglass-101 or DLA-34 in this project as they are both too heavy and time consuming.
DLA-34 should be a good choice as the backbone network real time object detector.
It's just opinion. I love your work.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.
@yjh0410 Hi!
You did a great job.
I did the benchmark test to find optimal backbone in terms of the execution time and imagenet top1 accuracy.
https://github.com/developer0hye/pytorch-backbone-benchmark
You can see DLA-34 is not heavy in comparison to other networks.
But you mentioned
CenterNet is an encoder-decoder network, but I won't consider Hourglass-101 or DLA-34 in this project as they are both too heavy and time consuming.
DLA-34 should be a good choice as the backbone network real time object detector.
I think that the neck of DLA-34 proposed in CenterNet paper is little heavy, but I guess we can get a good results if we use your simple neck structure
It's just opinion. I love your work.