caojunxu / ac-fpn Goto Github PK

Hello, I would like to ask, can AC-FPN be used in a single-stage target detection network? For example, YOLO V4. If yes, what needs to be modified? Hope you can answer, thank you！

PLEASE FOLLOW THESE INSTRUCTIONS BEFORE POSTING

1. Please thoroughly read README.md, INSTALL.md, GETTING_STARTED.md, and FAQ.md
2. Please search existing open and closed issues in case your issue has already been reported
3. Please try to debug the issue in case you can solve it on your own before posting

After following steps 1-3 above and agreeing to provide the detailed information requested below, you may continue with posting your issue

(Delete this line and the text above it.)

Expected results

What did you expect to see?

Actual results

What did you observe instead?

Detailed steps to reproduce

E.g.:

The command that you ran

System information

• Operating system: ?
• Compiler version: ?
• CUDA version: ?
• cuDNN version: ?
• NVIDIA driver version: ?
• GPU models (for all devices if they are not all the same): ?
• PYTHONPATH environment variable: ?
• python --version output: ?
• Anything else that seems relevant: ?

这个部分，你们是如何将N✖(W,H)的特征通过sigmiod和averagepool变成(1,W,H)的

Thank you very much for your open source. but I can't find where the AC-FPN code， could you tell me some detail？

我想问一下FPN上的注意力模块在哪里进行了实现，是在FPN.py吗，我在里面似乎没有找关于这些模块的搭建。感谢！

PLEASE FOLLOW THESE INSTRUCTIONS BEFORE POSTING

1. Please thoroughly read README.md, INSTALL.md, GETTING_STARTED.md, and FAQ.md
2. Please search existing open and closed issues in case your issue has already been reported
3. Please try to debug the issue in case you can solve it on your own before posting

After following steps 1-3 above and agreeing to provide the detailed information requested below, you may continue with posting your issue

(Delete this line and the text above it.)

Expected results

What did you expect to see?

Actual results

What did you observe instead?

Detailed steps to reproduce

E.g.:

The command that you ran

System information

• Operating system: ?
• Compiler version: ?
• CUDA version: ?
• cuDNN version: ?
• NVIDIA driver version: ?
• GPU models (for all devices if they are not all the same): ?
• PYTHONPATH environment variable: ?
• python --version output: ?
• Anything else that seems relevant: ?

Hi,
I have been trying to reproduce your paper using mmdetection, I refer to this repository and the version of paddlepaddle implemented by others. but when using the 1x learning schedule, the AP performance of Faster R-CNN w/Res50 has not been able to reach 38.5 (baseline is 37.4). I would like to ask, in the implementation of the CEM, what are the differences between it and DenseASPP besides Group Normalization?（I’m sorry I didn’t see the deformable convolution mentioned in the paper. Could you please point me?）

你好，感谢开源ACFPN，有2个问题想请教下

1. 下面这个地方is_upsample应该一直设置为True?最开始P5和C4融合时，就应该对P5进行降采样（我看backbone部分只修改了dilation，应该没有对C5的分辨率进行修改？）

2.论文中的CxAM和CnAM模块中的1x1卷积部分，是否对通道数进行了缩减了呢？nonlocal中是对计算通道数减半，减少了计算量，不知道你是否有意愿开源这两个模块呢？

期待你的回复，多谢！

请问AC-FPN的CEM模块有些类似DenseASPP，但是我看您的论文中并没有引用DenseASPP，这二者结构上或者功能上有什么区别呢？

hi nice work,I am wondering if the detectron implementation use dcn ,the paper mentioned,but i cannot find in fpn.py using dcn，also i cann't find it in the paper's graph too.And I want to ask if the CEM only used in p5?appreciated if you can reply.

Can not find CEM part in detectron/modeling/FPN.py, could you point out where it is?

A proper training steps as well as how to train on custom dataset will be beneficial? Plus a notebook will make the understanding easy