haotian/yolo_standard_libray

History

haotian 7ac785ddb3 first commit		2025-03-07 11:35:40 +08:00
..
gen_wts.py	first commit	2025-03-07 11:35:40 +08:00
holder.h	first commit	2025-03-07 11:35:40 +08:00
ibnnet.cpp	first commit	2025-03-07 11:35:40 +08:00
ibnnet.h	first commit	2025-03-07 11:35:40 +08:00
InferenceEngine.cpp	first commit	2025-03-07 11:35:40 +08:00
InferenceEngine.h	first commit	2025-03-07 11:35:40 +08:00
layers.cpp	first commit	2025-03-07 11:35:40 +08:00
layers.h	first commit	2025-03-07 11:35:40 +08:00
logging.h	first commit	2025-03-07 11:35:40 +08:00
main.cpp	first commit	2025-03-07 11:35:40 +08:00
README.md	first commit	2025-03-07 11:35:40 +08:00
utils.cpp	first commit	2025-03-07 11:35:40 +08:00
utils.h	first commit	2025-03-07 11:35:40 +08:00

README.md

IBN-Net

An implementation of IBN-Net, proposed in "Two at Once: Enhancing Learning and Generalization Capacities via IBN-Net", ECCV2018 by Xingang Pan, Ping Luo, Jianping Shi, Xiaoou Tang.

For the Pytorch implementation, you can refer to IBN-Net

Features

InstanceNorm2d
bottleneck_ibn
Resnet50-IBNA
Resnet50-IBNB
Multi-thread inference

How to Run

1. generate .wts
// for ibn-a
```
python gen_wts.py a
```
a file 'resnet50-ibna.wts' will be generated.

// for ibn-b
```
python gen_wts.py b
```
a file 'resnet50-ibnb.wts' will be generated.
1. cmake and make
```
mkdir build
cd build
cmake ..
make
```
1. build engine and run classification
// put resnet50-ibna.wts/resnet50-ibnb.wts into tensorrtx/ibnnet

// go to tensorrtx/ibnnet
```
./ibnnet -s  // serialize model to plan file
./ibnnet -d  // deserialize plan file and run inference
```