Skip to content

GitLab

Commit 522a602f authored by wangkx1's avatar wangkx1
Browse files

siton bug

parent abb99c90
Hide whitespace changes
Inline Side-by-side
Showing with 1041 additions and 0 deletions
configs/yolov5u/yolov5up6_m_300e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_300e_high.yml',
'_base_/yolov5up6_cspdarknet.yml',
'_base_/yolov5up6_reader.yml',
]
depth_mult: 0.67
width_mult: 0.75
log_iter: 100
snapshot_epoch: 10
weights: output/yolov5up6_m_300e_coco/model_final
TrainReader:
batch_size: 16 # default 8 gpus, total bs = 128
configs/yolov5u/yolov5up6_n_300e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_300e.yml',
'_base_/yolov5up6_cspdarknet.yml',
'_base_/yolov5up6_reader.yml',
]
depth_mult: 0.33
width_mult: 0.25
log_iter: 100
snapshot_epoch: 10
weights: output/yolov5up6_n_300e_coco/model_final
TrainReader:
batch_size: 16 # default 8 gpus, total bs = 128
configs/yolov5u/yolov5up6_s_300e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_300e.yml',
'_base_/yolov5up6_cspdarknet.yml',
'_base_/yolov5up6_reader.yml',
]
depth_mult: 0.33
width_mult: 0.50
log_iter: 100
snapshot_epoch: 10
weights: output/yolov5up6_s_300e_coco/model_final
TrainReader:
batch_size: 16 # default 8 gpus, total bs = 128
configs/yolov5u/yolov5up6_x_300e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_300e_high.yml',
'_base_/yolov5up6_cspdarknet.yml',
'_base_/yolov5up6_reader.yml',
]
depth_mult: 1.33
width_mult: 1.25
log_iter: 100
snapshot_epoch: 10
weights: output/yolov5up6_x_300e_coco/model_final
TrainReader:
batch_size: 8 # default 8 gpus, total bs = 64
configs/yolov6/README.md 0 → 100644
View file @ 522a602f
# YOLOv6 (YOLOv6 v3.0: A Full-Scale Reloading)
## 内容
- [模型库](#模型库)
- [使用说明](#使用说明)
- [速度测试](#速度测试)
- [引用](#引用)
## 模型库
### YOLOv6 on COCO
| 网络网络 | 输入尺寸 | 图片数/GPU | 学习率策略(蒸馏策略)| TRT-FP16-Latency(ms) | mAP | AP50 | Params(M) | FLOPs(G) | 下载链接 | 配置文件 |
| :------------- | :------- | :-------: | :--------------: | :---------: | :-----: |:-----: | :-----: |:-----: | :-------------: | :-----: |
| *YOLOv6-n | 640 | 16 | 300e(+300e) | 1.3 | 37.5 | 53.1 | 5.07 | 12.49 |[下载链接](https://paddledet.bj.bcebos.com/models/yolov6_n_300e_coco.pdparams) | [配置文件](./yolov6_n_300e_coco.yml) |
| *YOLOv6-s | 640 | 32 | 300e(+300e) | 2.7 | 44.8 | 61.7 | 20.18 | 49.36 |[下载链接](https://paddledet.bj.bcebos.com/models/yolov6_s_300e_coco.pdparams) | [配置文件](./yolov6_s_300e_coco.yml) |
| *YOLOv6-m | 640 | 32 | 300e(+300e) | 5.3 | 49.5 | 66.9 | 37.74 | 92.47 |[下载链接](https://paddledet.bj.bcebos.com/models/yolov6_m_300e_coco.pdparams) | [配置文件](./yolov6_m_300e_coco.yml) |
| *YOLOv6-l(silu) | 640 | 32 | 300e(+300e) | 9.5 | 52.2 | 70.2 | 59.66 | 149.4 |[下载链接](https://paddledet.bj.bcebos.com/models/yolov6_l_300e_coco.pdparams) | [配置文件](./yolov6_l_300e_coco.yml) |
**注意:**
- YOLOv6 模型为 v3.0 版本，暂未支持完全训练；训练使用COCO train2017作为训练集，mAP为在COCO val2017上的`mAP(IoU=0.5:0.95)`结果，且评估未使用`multi_label`等trick；
- YOLOv6 模型均需要先训练基础模型300epoch，再使用自蒸馏训练300epoch(增益1.0+ mAP)，且使用`dfl_loss`和设置`reg_max=16`；
- YOLOv6 n s 模型使用`EfficientRep`和`RepBiFPAN`，普通训练时采用`EffiDeHead`，蒸馏训练时采用`EffiDeHead_distill_ns`；
- YOLOv6 m l 模型使用`CSPBepBackbone`和`CSPRepBiFPAN`，普通训练时采用`EffiDeHead_fuseab`，蒸馏训练时采用`EffiDeHead`；
- YOLOv6 Params(M)和FLOPs(G)均为训练时所测；
- YOLOv6 l 模型默认激活函数为`silu`，其余 n s m 模型则默认为`relu`；
- YOLOv6 n 模型训练默认使用`siou loss`，其余s m l模型则默认使用`giou loss`；
- YOLOv6 模型训练过程中默认使用8 GPUs进行混合精度训练，默认每卡batch_size=32，其中YOLOv6 n模型每卡batch_size=16，自蒸馏训练阶段每卡batch_size和先前普通训练阶段一致，默认lr为0.01为8卡总batch_size的设置，如果**GPU卡数**或者每卡**batch size**发生改动，也不需要改动学习率，但为了保证高精度最好使用**总batch size大于64**的配置去训练；
- 模型推理耗时(ms)为TensorRT-FP16下测试的耗时，不包含数据预处理和模型输出后处理(NMS)的耗时。测试采用单卡Tesla T4 GPU，batch size=1，测试环境为**paddlepaddle-2.3.2**, **CUDA 11.2**, **CUDNN 8.2**, **GCC-8.2**, **TensorRT 8.0.3.4**，具体请参考[速度测试](#速度测试)。
- 如果你设置了`--run_benchmark=True`, 你首先需要安装以下依赖`pip install pynvml psutil GPUtil`。
### 部署模型
| 网络模型 | 输入尺寸 | 导出后的权重(w/o NMS) | ONNX(w/o NMS) |
| :-------- | :--------: | :---------------------: | :----------------: |
| yolov6-n | 640 | [(w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov6/yolov6_n_300e_coco_w_nms.zip) | [(w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov6/yolov6_n_300e_coco_wo_nms.zip) | [(w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov6/yolov6_n_300e_coco_w_nms.onnx) | [(w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov6/yolov6_n_300e_coco_wo_nms.onnx) |
| yolov6-s | 640 | [(w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov6/yolov6_s_300e_coco_w_nms.zip) | [(w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov6/yolov6_s_300e_coco_wo_nms.zip) | [(w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov6/yolov6_s_300e_coco_w_nms.onnx) | [(w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov6/yolov6_s_300e_coco_wo_nms.onnx) |
| yolov6-m | 640 | [(w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov6/yolov6_m_300e_coco_w_nms.zip) | [(w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov6/yolov6_m_300e_coco_wo_nms.zip) | [(w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov6/yolov6_m_300e_coco_w_nms.onnx) | [(w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov6/yolov6_m_300e_coco_wo_nms.onnx) |
| yolov6-l(silu) | 640 | [(w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov6/yolov6_l_300e_coco_w_nms.zip) | [(w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov6/yolov6_l_300e_coco_wo_nms.zip) | [(w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov6/yolov6_l_300e_coco_w_nms.onnx) | [(w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov6/yolov6_l_300e_coco_wo_nms.onnx) |
## 使用教程
### 0. **一键运行全流程**
将以下命令写在一个脚本文件里如```run.sh```，一键运行命令为：```sh run.sh```，也可命令行一句句去运行。
```bash
model_name=yolov6 # 可修改，如 ppyoloe
job_name=yolov6_s_300e_coco # 可修改，如 ppyoloe_plus_crn_s_80e_coco
config=configs/${model_name}/${job_name}.yml
log_dir=log_dir/${job_name}
# weights=https://bj.bcebos.com/v1/paddledet/models/${job_name}.pdparams
weights=output/${job_name}/model_final.pdparams
# 1.训练（单卡/多卡），加 --eval 表示边训边评估，加 --amp 表示混合精度训练
# CUDA_VISIBLE_DEVICES=0 python tools/train.py -c ${config} --eval --amp
python -m paddle.distributed.launch --log_dir=${log_dir} --gpus 0,1,2,3,4,5,6,7 tools/train.py -c ${config} --eval --amp
# 2.评估，加 --classwise 表示输出每一类mAP
CUDA_VISIBLE_DEVICES=0 python tools/eval.py -c ${config} -o weights=${weights} --classwise
# 3.预测 (单张图/图片文件夹）
CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ${config} -o weights=${weights} --infer_img=demo/000000014439_640x640.jpg --draw_threshold=0.5
# CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ${config} -o weights=${weights} --infer_dir=demo/ --draw_threshold=0.5
# 4.导出模型，以下3种模式选一种
## 普通导出，加trt表示用于trt加速，对NMS和silu激活函数提速明显
CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} # trt=True
## exclude_post_process去除后处理导出，返回和YOLOv5导出ONNX时相同格式的concat后的1个Tensor，是未缩放回原图的坐标+分类置信度
# CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} exclude_post_process=True # trt=True
## exclude_nms去除NMS导出，返回2个Tensor，是缩放回原图后的坐标和分类置信度
# CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} exclude_nms=True # trt=True
# 5.部署预测，注意不能使用 去除后处理 或 去除NMS 导出后的模型去预测
CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/${job_name} --image_file=demo/000000014439_640x640.jpg --device=GPU
# 6.部署测速，加 “--run_mode=trt_fp16” 表示在TensorRT FP16模式下测速，注意如需用到 trt_fp16 则必须为加 trt=True 导出的模型
CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/${job_name} --image_file=demo/000000014439_640x640.jpg --device=GPU --run_benchmark=True # --run_mode=trt_fp16
# 7.onnx导出，一般结合 exclude_post_process去除后处理导出的模型
paddle2onnx --model_dir output_inference/${job_name} --model_filename model.pdmodel --params_filename model.pdiparams --opset_version 12 --save_file ${job_name}.onnx
# 8.onnx trt测速
/usr/local/TensorRT-8.0.3.4/bin/trtexec --onnx=${job_name}.onnx --workspace=4096 --avgRuns=10 --shapes=input:1x3x640x640 --fp16
/usr/local/TensorRT-8.0.3.4/bin/trtexec --onnx=${job_name}.onnx --workspace=4096 --avgRuns=10 --shapes=input:1x3x640x640 --fp32
```
### 1. 训练
执行以下指令使用混合精度训练YOLOv6
```bash
python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 tools/train.py -c configs/yolov6/yolov6_s_300e_coco.yml --amp --eval
```
**注意:**
- `--amp`表示开启混合精度训练以避免显存溢出，`--eval`表示边训边验证。
### 2. 评估
执行以下命令在单个GPU上评估COCO val2017数据集
```bash
CUDA_VISIBLE_DEVICES=0 python tools/eval.py -c configs/yolov6/yolov6_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov6_s_300e_coco.pdparams
```
### 3. 推理
使用以下命令在单张GPU上预测图片，使用`--infer_img`推理单张图片以及使用`--infer_dir`推理文件中的所有图片。
```bash
# 推理单张图片
CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c configs/yolov6/yolov6_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov6_s_300e_coco.pdparams --infer_img=demo/000000014439_640x640.jpg
# 推理文件中的所有图片
CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c configs/yolov6/yolov6_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov6_s_300e_coco.pdparams --infer_dir=demo
```
### 4.导出模型
在GPU上推理部署或benchmark测速等需要通过`tools/export_model.py`导出模型。
当你**使用Paddle Inference但不使用TensorRT**时，运行以下的命令导出模型
```bash
python tools/export_model.py -c configs/yolov6/yolov6_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov6_s_300e_coco.pdparams
```
当你**使用Paddle Inference且使用TensorRT**时，需要指定`-o trt=True`来导出模型。
```bash
python tools/export_model.py -c configs/yolov6/yolov6_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov6_s_300e_coco.pdparams trt=True
```
如果你想将YOLOv6模型导出为**ONNX格式**，参考
[PaddleDetection模型导出为ONNX格式教程](../../deploy/EXPORT_ONNX_MODEL.md)，运行以下命令：
```bash
# 导出推理模型
python tools/export_model.py -c configs/yolov6/yolov6_s_300e_coco.yml --output_dir=output_inference -o weights=https://paddledet.bj.bcebos.com/models/yolov6_s_300e_coco.pdparams
# 安装paddle2onnx
pip install paddle2onnx
# 转换成onnx格式
paddle2onnx --model_dir output_inference/yolov6_s_300e_coco --model_filename model.pdmodel --params_filename model.pdiparams --opset_version 11 --save_file yolov6_s_300e_coco.onnx
```
**注意：** ONNX模型目前只支持batch_size=1
### 5.推理部署
YOLOv6可以使用以下方式进行部署：
- Paddle Inference [Python](../../deploy/python) & [C++](../../deploy/cpp)
- [Paddle-TensorRT](../../deploy/TENSOR_RT.md)
- [PaddleServing](https://github.com/PaddlePaddle/Serving)
- [PaddleSlim模型量化](../slim)
运行以下命令导出模型
```bash
python tools/export_model.py -c configs/yolov6/yolov6_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov6_s_300e_coco.pdparams trt=True
```
**注意：**
- trt=True表示**使用Paddle Inference且使用TensorRT**进行测速，速度会更快，默认不加即为False，表示**使用Paddle Inference但不使用TensorRT**进行测速。
- 如果是使用Paddle Inference在TensorRT FP16模式下部署，需要参考[Paddle Inference文档](https://www.paddlepaddle.org.cn/inference/master/user_guides/download_lib.html#python)，下载并安装与你的CUDA, CUDNN和TensorRT相应的wheel包。
#### 5.1.Python部署
`deploy/python/infer.py`使用上述导出后的Paddle Inference模型用于推理和benchnark测速，如果设置了`--run_benchmark=True`, 首先需要安装以下依赖`pip install pynvml psutil GPUtil`。
```bash
# Python部署推理单张图片
python deploy/python/infer.py --model_dir=output_inference/yolov6_s_300e_coco --image_file=demo/000000014439_640x640.jpg --device=gpu
# 推理文件夹下的所有图片
python deploy/python/infer.py --model_dir=output_inference/yolov6_s_300e_coco --image_dir=demo/ --device=gpu
```
#### 5.2. C++部署
`deploy/cpp/build/main`使用上述导出后的Paddle Inference模型用于C++推理部署, 首先按照[docs](../../deploy/cpp/docs)编译安装环境。
```bash
# C++部署推理单张图片
./deploy/cpp/build/main --model_dir=output_inference/yolov6_s_300e_coco/ --image_file=demo/000000014439_640x640.jpg --run_mode=paddle --device=GPU --threshold=0.5 --output_dir=cpp_infer_output/yolov6_s_300e_coco
```
## 速度测试
为了公平起见，在[模型库](#模型库)中的速度测试结果均为不包含数据预处理和模型输出后处理(NMS)的数据(与[YOLOv4(AlexyAB)](https://github.com/AlexeyAB/darknet)测试方法一致)，需要在导出模型时指定`-o exclude_nms=True`。测速需设置`--run_benchmark=True`, 首先需要安装以下依赖`pip install pynvml psutil GPUtil`。
**使用Paddle Inference但不使用TensorRT**进行测速，执行以下命令：
```bash
# 导出模型
python tools/export_model.py -c configs/yolov6/yolov6_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov6_s_300e_coco.pdparams exclude_nms=True
# 速度测试，使用run_benchmark=True
python deploy/python/infer.py --model_dir=output_inference/yolov6_s_300e_coco --image_file=demo/000000014439_640x640.jpg --run_mode=paddle --device=gpu --run_benchmark=True
```
**使用Paddle Inference且使用TensorRT**进行测速，执行以下命令：
```bash
# 导出模型，使用trt=True
python tools/export_model.py -c configs/yolov6/yolov6_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov6_s_300e_coco.pdparams exclude_nms=True trt=True
# 速度测试，使用run_benchmark=True
python deploy/python/infer.py --model_dir=output_inference/yolov6_s_300e_coco --image_file=demo/000000014439_640x640.jpg --device=gpu --run_benchmark=True
# tensorRT-FP32测速
python deploy/python/infer.py --model_dir=output_inference/yolov6_s_300e_coco --image_file=demo/000000014439_640x640.jpg --device=gpu --run_benchmark=True --run_mode=trt_fp32
# tensorRT-FP16测速
python deploy/python/infer.py --model_dir=output_inference/yolov6_s_300e_coco --image_file=demo/000000014439_640x640.jpg --device=gpu --run_benchmark=True --run_mode=trt_fp16
```
**注意:**
- 导出模型时指定`-o exclude_nms=True`仅作为测速时用，这样导出的模型其推理部署预测的结果不是最终检出框的结果。
- [模型库](#模型库)中的速度测试结果为tensorRT-FP16测速后的最快速度，为不包含数据预处理和模型输出后处理(NMS)的耗时。
## 引用
```
@article{li2022yolov6,
title={YOLOv6: A Single-Stage Object Detection Framework for Industrial Applications},
journal={arXiv preprint arXiv:2207.02696},
year={2022}
}
@article{li2022yolov6,
title={YOLOv6 v3.0: A Full-Scale Reloading},
journal={arXiv preprint arXiv:2301.05586},
year={2023}
}
```
configs/yolov6/_base_/optimizer_300e.yml 0 → 100644
View file @ 522a602f
epoch: 300
LearningRate:
base_lr: 0.01
schedulers:
- !YOLOv5LRDecay
max_epochs: 300
min_lr_ratio: 0.01
- !ExpWarmup
epochs: 3
OptimizerBuilder:
optimizer:
type: Momentum
momentum: 0.937
use_nesterov: True
regularizer:
factor: 0.0005
type: L2
configs/yolov6/_base_/optimizer_300e_fintune.yml 0 → 100644
View file @ 522a602f
epoch: 300
LearningRate:
base_lr: 0.0032
schedulers:
- !YOLOv5LRDecay
max_epochs: 300
min_lr_ratio: 0.12
- !ExpWarmup
epochs: 2
OptimizerBuilder:
optimizer:
type: Momentum
momentum: 0.843
use_nesterov: True
regularizer:
factor: 0.00036
type: L2
configs/yolov6/_base_/yolov6_cspbep.yml 0 → 100644
View file @ 522a602f
architecture: YOLOv6
norm_type: sync_bn
use_ema: True
ema_decay: 0.9999
ema_decay_type: "exponential"
find_unused_parameters: True
act: 'relu' # 'silu' in L model
training_mode: "repvgg"
self_distill: False
depth_mult: 1.0 # default: L model
width_mult: 1.0
YOLOv6:
backbone: CSPBepBackbone
neck: CSPRepBiFPAN
yolo_head: EffiDeHead_fuseab
post_process: ~
CSPBepBackbone:
arch: 'P5'
return_idx: [2, 3, 4]
csp_e: 0.5
fuse_P2: True # add P2 and return 4 layers
cspsppf: False
CSPRepBiFPAN:
csp_e: 0.5
EffiDeHead_fuseab:
fpn_strides: [8, 16, 32]
grid_cell_scale: 5.0
grid_cell_offset: 0.5
reg_max: 16
use_dfl: True # in m/l version
static_assigner_epoch: 3 # warmup_epoch
loss_weight: {cls: 1.0, iou: 2.5, dfl: 0.5, cwd: 10.0}
iou_type: 'giou' # 'siou' in n/t version, 'giou' in s/m/l version
distill_weight: {cls: 1.0, dfl: 1.0}
static_assigner:
name: ATSSAssigner
topk: 9
assigner:
name: TaskAlignedAssigner
topk: 13
alpha: 1.0
beta: 6.0
nms:
name: MultiClassNMS
nms_top_k: 3000
keep_top_k: 300
score_threshold: 0.03
nms_threshold: 0.65
EffiDeHead:
fpn_strides: [8, 16, 32]
grid_cell_scale: 5.0
grid_cell_offset: 0.5
reg_max: 16
use_dfl: True # in m/l version
static_assigner_epoch: 3 # warmup_epoch
loss_weight: {cls: 1.0, iou: 2.5, dfl: 0.5, cwd: 10.0}
iou_type: 'giou' # 'siou' in n/t version, 'giou' in s/m/l version
distill_weight: {cls: 1.0, dfl: 1.0}
static_assigner:
name: ATSSAssigner
topk: 9
assigner:
name: TaskAlignedAssigner
topk: 13
alpha: 1.0
beta: 6.0
nms:
name: MultiClassNMS
nms_top_k: 3000
keep_top_k: 300
score_threshold: 0.03
nms_threshold: 0.65
configs/yolov6/_base_/yolov6_efficientrep.yml 0 → 100644
View file @ 522a602f
architecture: YOLOv6
norm_type: sync_bn
use_ema: True
ema_decay: 0.9999
ema_decay_type: "exponential"
find_unused_parameters: True
act: 'relu'
training_mode: "repvgg"
self_distill: False
depth_mult: 0.33 # default: s model
width_mult: 0.50
YOLOv6:
backbone: EfficientRep
neck: RepBiFPAN
yolo_head: EffiDeHead
post_process: ~
EfficientRep:
arch: 'P5'
return_idx: [2, 3, 4]
fuse_P2: True
cspsppf: True
# use default config
# RepBiFPAN:
EffiDeHead:
fpn_strides: [8, 16, 32]
grid_cell_scale: 5.0
grid_cell_offset: 0.5
reg_max: 0
use_dfl: False
static_assigner_epoch: 4 # warmup_epoch
loss_weight: {cls: 1.0, iou: 2.5}
iou_type: 'siou' # 'siou' in n version, 'giou' in s/m/l version
static_assigner:
name: ATSSAssigner
topk: 9
assigner:
name: TaskAlignedAssigner
topk: 13
alpha: 1.0
beta: 6.0
nms:
name: MultiClassNMS
nms_top_k: 3000
keep_top_k: 300
score_threshold: 0.03
nms_threshold: 0.65
EffiDeHead_distill_ns:
fpn_strides: [8, 16, 32]
grid_cell_scale: 5.0
grid_cell_offset: 0.5
reg_max: 16
use_dfl: True
static_assigner_epoch: 4 # warmup_epoch
loss_weight: {cls: 1.0, iou: 2.5}
iou_type: 'siou' # 'siou' in n version, 'giou' in s/m/l version
static_assigner:
name: ATSSAssigner
topk: 9
assigner:
name: TaskAlignedAssigner
topk: 13
alpha: 1.0
beta: 6.0
nms:
name: MultiClassNMS
nms_top_k: 3000
keep_top_k: 300
score_threshold: 0.03
nms_threshold: 0.65
configs/yolov6/_base_/yolov6_reader.yml 0 → 100644
View file @ 522a602f
input_height: &input_height 640
input_width: &input_width 640
input_size: &input_size [*input_height, *input_width]
mosaic_epoch: &mosaic_epoch 300
worker_num: 4
TrainReader:
sample_transforms:
- DecodeNormResize: {target_size: *input_size, mosaic: True}
- MosaicPerspective:
mosaic_prob: 1.0
target_size: *input_size
degrees: 0.0
translate: 0.1
scale: 0.5
shear: 0.0
mixup_prob: 0.0
- RandomHSV: {hgain: 0.015, sgain: 0.7, vgain: 0.4}
- RandomFlip: {}
batch_transforms:
- NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
- Permute: {}
- PadGT: {}
batch_size: 32
shuffle: True
drop_last: True
use_shared_memory: True
collate_batch: True
mosaic_epoch: *mosaic_epoch
EvalReader:
sample_transforms:
- Decode: {}
- Resize: {target_size: *input_size, keep_ratio: True, interp: 1}
- Pad: {size: *input_size, fill_value: [114., 114., 114.]}
- NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
- Permute: {}
batch_size: 1
TestReader:
inputs_def:
image_shape: [3, *input_height, *input_width]
sample_transforms:
- Decode: {}
- Resize: {target_size: *input_size, keep_ratio: True, interp: 1}
- Pad: {size: *input_size, fill_value: [114., 114., 114.]}
- NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
- Permute: {}
batch_size: 1
fuse_normalize: False
configs/yolov6/_base_/yolov6_reader_fintune.yml 0 → 100644
View file @ 522a602f
input_height: &input_height 640
input_width: &input_width 640
input_size: &input_size [*input_height, *input_width]
mosaic_epoch: &mosaic_epoch 300
worker_num: 4
TrainReader:
sample_transforms:
- DecodeNormResize: {target_size: *input_size, mosaic: True}
- MosaicPerspective:
mosaic_prob: 1.0
target_size: *input_size
degrees: 0.373
translate: 0.245
scale: 0.898
shear: 0.602
mixup_prob: 0.243
- RandomHSV: {hgain: 0.0138, sgain: 0.664, vgain: 0.464}
- RandomFlip: {}
batch_transforms:
- NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
- Permute: {}
- PadGT: {}
batch_size: 32
shuffle: True
drop_last: True
use_shared_memory: True
collate_batch: True
mosaic_epoch: *mosaic_epoch
EvalReader:
sample_transforms:
- Decode: {}
- Resize: {target_size: *input_size, keep_ratio: True, interp: 1}
- Pad: {size: *input_size, fill_value: [114., 114., 114.]}
- NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
- Permute: {}
batch_size: 1
TestReader:
inputs_def:
image_shape: [3, *input_height, *input_width]
sample_transforms:
- Decode: {}
- Resize: {target_size: *input_size, keep_ratio: True, interp: 1}
- Pad: {size: *input_size, fill_value: [114., 114., 114.]}
- NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
- Permute: {}
batch_size: 1
fuse_normalize: False
configs/yolov6/_base_/yolov6_reader_high_aug.yml 0 → 100644
View file @ 522a602f
input_height: &input_height 640
input_width: &input_width 640
input_size: &input_size [*input_height, *input_width]
mosaic_epoch: &mosaic_epoch 300
worker_num: 4
TrainReader:
sample_transforms:
- DecodeNormResize: {target_size: *input_size, mosaic: True}
- MosaicPerspective:
mosaic_prob: 1.0
target_size: *input_size
degrees: 0.0
translate: 0.1
scale: 0.9 #
shear: 0.0
mixup_prob: 0.1 #
- RandomHSV: {hgain: 0.015, sgain: 0.7, vgain: 0.4}
- RandomFlip: {}
batch_transforms:
- NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
- Permute: {}
- PadGT: {}
batch_size: 32
shuffle: True
drop_last: True
use_shared_memory: True
collate_batch: True
mosaic_epoch: *mosaic_epoch
EvalReader:
sample_transforms:
- Decode: {}
- Resize: {target_size: *input_size, keep_ratio: True, interp: 1}
- Pad: {size: *input_size, fill_value: [114., 114., 114.]}
- NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
- Permute: {}
batch_size: 1
TestReader:
inputs_def:
image_shape: [3, *input_height, *input_width]
sample_transforms:
- Decode: {}
- Resize: {target_size: *input_size, keep_ratio: True, interp: 1}
- Pad: {size: *input_size, fill_value: [114., 114., 114.]}
- NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
- Permute: {}
batch_size: 1
fuse_normalize: False
configs/yolov6/yolov6_l_300e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_300e.yml',
'_base_/yolov6_cspbep.yml',
'_base_/yolov6_reader_high_aug.yml',
]
depth_mult: 1.0
width_mult: 1.0
log_iter: 20
snapshot_epoch: 10
weights: output/yolov6_l_300e_coco/model_final
### reader config
TrainReader:
batch_size: 32 # default 8 gpus, total bs = 256
EvalReader:
batch_size: 1
### model config
act: 'silu'
training_mode: "conv_silu" # Note: L use silu
YOLOv6:
backbone: CSPBepBackbone
neck: CSPRepBiFPAN
yolo_head: EffiDeHead_fuseab
post_process: ~
EffiDeHead_fuseab:
reg_max: 16
use_dfl: True
iou_type: 'giou'
loss_weight: {cls: 1.0, iou: 2.5, dfl: 0.5, cwd: 10.0}
distill_weight: {cls: 2.0, dfl: 1.0} # 2:1 in L-relu version, will not work default (self_distill=False)
CSPBepBackbone:
csp_e: 0.50
CSPRepBiFPAN:
csp_e: 0.50
### distill config
## Step 1: Training the base model, get about 51.8 mAP
## Step 2: Self-distillation training, get about 52.8 mAP
YOLOv6:
backbone: CSPBepBackbone
neck: CSPRepBiFPAN
yolo_head: EffiDeHead
post_process: ~
EffiDeHead:
reg_max: 16
use_dfl: True
## Please cancel the following comment and train again:
# self_distill: True
# pretrain_weights: output/yolov6_l_300e_coco/model_final.pdparams
# save_dir: output_distill
# weights: output_distill/yolov6_l_300e_coco/model_final
configs/yolov6/yolov6_m_300e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_300e.yml',
'_base_/yolov6_cspbep.yml',
'_base_/yolov6_reader_high_aug.yml',
]
depth_mult: 0.60
width_mult: 0.75
log_iter: 20
snapshot_epoch: 10
weights: output/yolov6_m_300e_coco/model_final
### reader config
TrainReader:
batch_size: 32 # default 8 gpus, total bs = 256
EvalReader:
batch_size: 1
### model config
act: 'relu'
training_mode: "repvgg"
YOLOv6:
backbone: CSPBepBackbone
neck: CSPRepBiFPAN
yolo_head: EffiDeHead_fuseab
post_process: ~
EffiDeHead_fuseab:
reg_max: 16
use_dfl: True
static_assigner_epoch: 4
iou_type: 'giou'
loss_weight: {cls: 1.0, iou: 2.5, dfl: 0.5, cwd: 10.0}
distill_weight: {cls: 1.0, dfl: 1.0} # 1:1 , will not work default (self_distill=False)
CSPBepBackbone:
csp_e: 0.67
CSPRepBiFPAN:
csp_e: 0.67
### distill config
## Step 1: Training the base model, get about 49.1 mAP
## Step 2: Self-distillation training, get about 50.0 mAP
YOLOv6:
backbone: CSPBepBackbone
neck: CSPRepBiFPAN
yolo_head: EffiDeHead
post_process: ~
EffiDeHead:
reg_max: 16
use_dfl: True
## Please cancel the following comment and train again:
# self_distill: True
# pretrain_weights: output/yolov6_m_300e_coco/model_final.pdparams
# save_dir: output_distill
# weights: output_distill/yolov6_m_300e_coco/model_final
configs/yolov6/yolov6_n_300e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_300e.yml',
'_base_/yolov6_efficientrep.yml',
'_base_/yolov6_reader.yml',
]
depth_mult: 0.33
width_mult: 0.25
log_iter: 20
snapshot_epoch: 10
weights: output/yolov6_n_300e_coco/model_final
### reader config
TrainReader:
batch_size: 16 # default 8 gpus, total bs = 128
EvalReader:
batch_size: 1
### model config
act: 'relu'
training_mode: "repvgg"
YOLOv6:
backbone: EfficientRep
neck: RepBiFPAN
yolo_head: EffiDeHead
post_process: ~
EffiDeHead:
reg_max: 0
use_dfl: False # False in n/s
loss_weight: {cls: 1.0, iou: 2.5}
iou_type: 'siou' # only in n/t version
### distill config
## Step 1: Training the base model, get about 37.0 mAP
## Step 2: Self-distillation training, get about 37.5 mAP
YOLOv6:
backbone: EfficientRep
neck: RepBiFPAN
yolo_head: EffiDeHead_distill_ns
post_process: ~
EffiDeHead_distill_ns:
reg_max: 16
use_dfl: True
## Please cancel the following comment and train again:
# self_distill: True
# pretrain_weights: output/yolov6_n_300e_coco/model_final.pdparams
# save_dir: output_distill
# weights: output_distill/yolov6_n_300e_coco/model_final
configs/yolov6/yolov6_s_300e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_300e.yml',
'_base_/yolov6_efficientrep.yml',
'_base_/yolov6_reader.yml',
]
depth_mult: 0.33
width_mult: 0.50
log_iter: 20
snapshot_epoch: 10
weights: output/yolov6_s_300e_coco/model_final
### reader config
TrainReader:
batch_size: 32 # default 8 gpus, total bs = 256
EvalReader:
batch_size: 1
### model config
act: 'relu'
training_mode: "repvgg"
YOLOv6:
backbone: EfficientRep
neck: RepBiFPAN
yolo_head: EffiDeHead
post_process: ~
EffiDeHead:
reg_max: 0
use_dfl: False # False in n/s
loss_weight: {cls: 1.0, iou: 2.5}
iou_type: 'giou'
### distill config
## Step 1: Training the base model, get about 44.3 mAP
## Step 2: Self-distillation training, get about 45.0 mAP
YOLOv6:
backbone: EfficientRep
neck: RepBiFPAN
yolo_head: EffiDeHead_distill_ns
post_process: ~
EffiDeHead_distill_ns:
reg_max: 16
use_dfl: True
## Please cancel the following comment and train again:
# self_distill: True
# pretrain_weights: output/yolov6_s_300e_coco/model_final.pdparams
# save_dir: output_distill
# weights: output_distill/yolov6_s_300e_coco/model_final
configs/yolov6_seg/README.md 0 → 100644
View file @ 522a602f
# YOLOv6 Instance segmentation
## 模型库
### 实例分割模型
| 网络网络 | 输入尺寸 | 图片数/GPU | 学习率策略 | 模型推理耗时(ms) | box AP | mask AP | Params(M) | FLOPs(G) | 下载链接 | 配置文件 |
| :------------- | :------- | :-------: | :------: | :------------: | :---------------------: | :----------------: |:---------: | :------: |:---------------: |:-----: |
| *YOLOv6-n | 640 | 8 | 300e | - | 36.3 | - | - | - | [下载链接](https://paddledet.bj.bcebos.com/models/yolov6_seg_n_300e_coco.pdparams) | [配置文件](./yolov6_seg_n_300e_coco.yml) |
| *YOLOv6-s | 640 | 8 | 300e | - | 44.0 | - | - | - | [下载链接](https://paddledet.bj.bcebos.com/models/yolov6_seg_s_300e_coco.pdparams) | [配置文件](./yolov6_seg_s_300e_coco.yml) |
| *YOLOv6-m | 640 | 8 | 300e | - | 48.3 | - | - | - | [下载链接](https://paddledet.bj.bcebos.com/models/yolov6_seg_m_300e_coco.pdparams) | [配置文件](./yolov6_seg_m_300e_coco.yml) |
| *YOLOv6-l | 640 | 8 | 300e | - | 50.8 | - | - | - | [下载链接](https://paddledet.bj.bcebos.com/models/yolov6_seg_l_300e_coco.pdparams) | [配置文件](./yolov6_seg_l_300e_coco.yml) |
| *YOLOv6-x | 640 | 8 | 300e | - | 52.1 | - | - | - | [下载链接](https://paddledet.bj.bcebos.com/models/yolov6_seg_x_300e_coco.pdparams) | [配置文件](./yolov6_seg_x_300e_coco.yml) |
configs/yolov6_seg/_base_/optimizer_300e.yml 0 → 100644
View file @ 522a602f
epoch: 300
LearningRate:
base_lr: 0.01
schedulers:
- !YOLOv5LRDecay
max_epochs: 300
min_lr_ratio: 0.01
- !ExpWarmup
epochs: 3
OptimizerBuilder:
optimizer:
type: Momentum
momentum: 0.937
use_nesterov: True
regularizer:
factor: 0.0005
type: L2
configs/yolov6_seg/_base_/yolov6_seg_cspbep.yml 0 → 100644
View file @ 522a602f
architecture: YOLOv6
norm_type: sync_bn
use_ema: True
ema_decay: 0.9999
ema_decay_type: "exponential"
find_unused_parameters: True
with_mask: True
act: 'relu' # 'silu' in L model
training_mode: "repvgg"
depth_mult: 1.0 # default: L model
width_mult: 1.0
YOLOv6:
backbone: CSPBepBackbone
neck: CSPRepBiFPAN
yolo_head: EffiDeInsHead
post_process: ~
CSPBepBackbone:
arch: 'P5'
return_idx: [2, 3, 4]
csp_e: 0.5
fuse_P2: True # add P2 and return 4 layers
cspsppf: False
CSPRepBiFPAN:
csp_e: 0.5
EffiDeInsHead:
fpn_strides: [8, 16, 32]
grid_cell_scale: 5.0
grid_cell_offset: 0.5
reg_max: 16
use_dfl: True # in m/l version
static_assigner_epoch: 0 # warmup_epoch
loss_weight: {cls: 1.0, iou: 2.5, dfl: 0.5}
iou_type: 'giou' # 'siou' in n version, 'giou' in s/m/l/x version
static_assigner:
name: ATSSAssigner
topk: 9
assigner:
name: TaskAlignedAssigner
topk: 13
alpha: 1.0
beta: 6.0
nms:
name: MultiClassNMS
nms_top_k: 3000
keep_top_k: 300
score_threshold: 0.03
nms_threshold: 0.65
return_index: True
configs/yolov6_seg/_base_/yolov6_seg_efficientrep.yml 0 → 100644
View file @ 522a602f
architecture: YOLOv6
norm_type: sync_bn
use_ema: True
ema_decay: 0.9999
ema_decay_type: "exponential"
find_unused_parameters: True
with_mask: True
act: 'relu'
training_mode: "repvgg"
depth_mult: 0.33 # default: s model
width_mult: 0.50
YOLOv6:
backbone: EfficientRep
neck: RepBiFPAN
yolo_head: EffiDeInsHead
post_process: ~
EfficientRep:
arch: 'P5'
return_idx: [2, 3, 4]
fuse_P2: True
cspsppf: True
# use default config
# RepBiFPAN:
EffiDeInsHead:
fpn_strides: [8, 16, 32]
grid_cell_scale: 5.0
grid_cell_offset: 0.5
reg_max: 16 #
use_dfl: True #
static_assigner_epoch: 0 # warmup_epoch
loss_weight: {cls: 1.0, iou: 2.5}
iou_type: 'giou' # 'siou' in n version, 'giou' in s/m/l/x version
static_assigner:
name: ATSSAssigner
topk: 9
assigner:
name: TaskAlignedAssigner
topk: 13
alpha: 1.0
beta: 6.0
nms:
name: MultiClassNMS
nms_top_k: 3000
keep_top_k: 300
score_threshold: 0.03
nms_threshold: 0.65
return_index: True
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment