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 738 additions and 0 deletions
configs/yolov7/_base_/yolov7_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, mixup_prob: 0.15, paste_in_prob: 0.15, translate: 0.2, scale: 0.9}
- RandomHSV: {hgain: 0.015, sgain: 0.7, vgain: 0.4}
- RandomFlip: {}
- BboxXYXY2XYWH: {}
- NormalizeBox: {}
- NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
- RGBReverse: {} # bgr->rgb
- Permute: {}
batch_size: 32
shuffle: True
drop_last: False
use_shared_memory: False
collate_batch: False
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, 640, 640]
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/yolov7/_base_/yolov7_tiny_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, mixup_prob: 0.05, paste_in_prob: 0.05, translate: 0.1, scale: 0.5}
- RandomHSV: {hgain: 0.015, sgain: 0.7, vgain: 0.4}
- RandomFlip: {}
- BboxXYXY2XYWH: {}
- NormalizeBox: {}
- NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
- RGBReverse: {} # bgr->rgb
- Permute: {}
batch_size: 32
shuffle: True
drop_last: False
use_shared_memory: False
collate_batch: False
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, 640, 640]
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/yolov7/_base_/yolov7p6_elannet.yml 0 → 100644
View file @ 522a602f
architecture: YOLOv7
# norm_type: sync_bn
use_ema: True
ema_decay: 0.9999
ema_decay_type: "exponential"
act: silu
find_unused_parameters: True
depth_mult: 1.0 # no use in YOLOv7
width_mult: 1.0
arch: 'W6'
use_aux: True # set False if compute params or flops
use_implicit: False # default False here, True in paper
YOLOv7:
backbone: ELANNet
neck: ELANFPNP6
yolo_head: YOLOv7Head
post_process: ~
ELANNet:
return_idx: [2, 3, 4, 5]
depthwise: false
ELANFPNP6:
depthwise: false
YOLOv7Head:
anchors: [[19, 27], [44, 40], [38, 94],
[96, 68], [86, 152], [180, 137],
[140, 301], [303, 264], [238, 542],
[436, 615], [739, 380], [925, 792]]
anchor_masks: [[0, 1, 2], [3, 4, 5], [6, 7, 8], [9, 10, 11]]
stride: [8, 16, 32, 64]
loss: YOLOv7Loss
nms:
name: MultiClassNMS
nms_top_k: 10000
keep_top_k: 300
score_threshold: 0.001
nms_threshold: 0.7
YOLOv7Loss:
downsample_ratios: [8, 16, 32, 64]
balance: [4.0, 1.0, 0.25, 0.06]
box_weight: 0.05
cls_weght: 0.3
obj_weight: 0.7
configs/yolov7/_base_/yolov7p6_reader.yml 0 → 100644
View file @ 522a602f
input_height: &input_height 1280
input_width: &input_width 1280
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, mixup_prob: 0.15, paste_in_prob: 0.15, translate: 0.2, scale: 0.9}
- RandomHSV: {hgain: 0.015, sgain: 0.7, vgain: 0.4}
- RandomFlip: {}
- BboxXYXY2XYWH: {}
- NormalizeBox: {}
batch_transforms:
- NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
- Permute: {}
- PadGT: {}
batch_size: 16
shuffle: True
drop_last: False
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: 2
TestReader:
inputs_def:
image_shape: [3, 1280, 1280]
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/yolov7/yolov7_l_300e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_300e.yml',
'_base_/yolov7_elannet.yml',
'_base_/yolov7_reader.yml',
]
log_iter: 100
snapshot_epoch: 10
weights: output/yolov7_l_300e_coco/model_final
arch: L
TrainReader:
batch_size: 16
configs/yolov7/yolov7_tiny_300e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_300e_tiny.yml',
'_base_/yolov7_elannet.yml',
'_base_/yolov7_tiny_reader.yml',
]
log_iter: 100
snapshot_epoch: 10
weights: output/yolov7_tiny_300e_coco/model_final
arch: tiny
act: LeakyReLU
TrainReader:
batch_size: 16
YOLOv7Loss:
box_weight: 0.05
cls_weght: 0.5
obj_weight: 1.0
configs/yolov7/yolov7_tiny_320_300e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'yolov7_tiny_300e_coco.yml',
]
log_iter: 100
snapshot_epoch: 10
weights: output/yolov7_tiny_320_300e_coco/model_final
EvalReader:
sample_transforms:
- Decode: {}
- Resize: {target_size: [320, 320], keep_ratio: True, interp: 1}
- Pad: {size: [320, 320], 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, 320, 320]
sample_transforms:
- Decode: {}
- Resize: {target_size: [320, 320], keep_ratio: True, interp: 1}
- Pad: {size: [320, 320], 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/yolov7/yolov7_tiny_416_300e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'yolov7_tiny_300e_coco.yml',
]
log_iter: 100
snapshot_epoch: 10
weights: output/yolov7_tiny_416_300e_coco/model_final
EvalReader:
sample_transforms:
- Decode: {}
- Resize: {target_size: [416, 416], keep_ratio: True, interp: 1}
- Pad: {size: [416, 416], 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, 416, 416]
sample_transforms:
- Decode: {}
- Resize: {target_size: [416, 416], keep_ratio: True, interp: 1}
- Pad: {size: [416, 416], 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/yolov7/yolov7_x_300e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_300e.yml',
'_base_/yolov7_elannet.yml',
'_base_/yolov7_reader.yml',
]
log_iter: 100
snapshot_epoch: 10
weights: output/yolov7_x_300e_coco/model_final
arch: X
TrainReader:
batch_size: 16
configs/yolov7/yolov7p6_d6_300e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_300e_p6.yml',
'_base_/yolov7p6_elannet.yml',
'_base_/yolov7p6_reader.yml',
]
log_iter: 100
snapshot_epoch: 10
weights: output/yolov7p6_d6_300e_coco/model_final
arch: D6
use_aux: True
TrainReader:
batch_size: 8
configs/yolov7/yolov7p6_e6_300e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_300e_p6.yml',
'_base_/yolov7p6_elannet.yml',
'_base_/yolov7p6_reader.yml',
]
log_iter: 100
snapshot_epoch: 10
weights: output/yolov7p6_e6_300e_coco/model_final
arch: E6
use_aux: True
TrainReader:
batch_size: 10
configs/yolov7/yolov7p6_e6e_300e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_300e_p6.yml',
'_base_/yolov7p6_elannet.yml',
'_base_/yolov7p6_reader.yml',
]
log_iter: 100
snapshot_epoch: 10
weights: output/yolov7p6_e6e_300e_coco/model_final
arch: E6E
use_aux: True
TrainReader:
batch_size: 6
configs/yolov7/yolov7p6_w6_300e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_300e_p6.yml',
'_base_/yolov7p6_elannet.yml',
'_base_/yolov7p6_reader.yml',
]
log_iter: 100
snapshot_epoch: 10
weights: output/yolov7p6_w6_300e_coco/model_final
arch: W6
use_aux: True
TrainReader:
batch_size: 16
configs/yolov7u/README.md 0 → 100644
View file @ 522a602f
# YOLOv7u
#### YOLOv7u 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) | mAP<sup>val<br>0.5:0.95 | mAP<sup>val<br>0.5 | Params(M) | FLOPs(G) | 下载链接 | 配置文件 |
| :------------- | :------- | :-------: | :------: | :------------: | :---------------------: | :----------------: |:---------: | :------: |:---------------: |:-----: |
| YOLOv7u-L | 640 | 16 | 300e | 9.0 | 52.1 | 68.8 | 43.59 | 130.10 |[下载链接](https://paddledet.bj.bcebos.com/models/yolov7u_l_300e_coco.pdparams) | [配置文件](./yolov7u_l_300e_coco.yml) |
**注意:**
- YOLOv7u 模型表示YOLOv7结构使用YOLOv8的head和loss，并结合YOLOR的ImplicitA和ImplicitM，是Anchor Free的检测方案；
- YOLOv7u 模型训练使用COCO train2017作为训练集，Box AP为在COCO val2017上的`mAP(IoU=0.5:0.95)`结果；
- 使用教程可参照[YOLOv7](../yolov7)；
configs/yolov7u/_base_/optimizer_300e.yml 0 → 100644
View file @ 522a602f
epoch: 300
LearningRate:
base_lr: 0.01
schedulers:
- !YOLOv5LRDecay
max_epochs: 300
min_lr_ratio: 0.1
- !ExpWarmup
epochs: 3
OptimizerBuilder:
optimizer:
type: Momentum
momentum: 0.937
use_nesterov: True
regularizer:
factor: 0.001
type: L2
configs/yolov7u/_base_/yolov7u_elannet.yml 0 → 100644
View file @ 522a602f
architecture: YOLOv7
norm_type: sync_bn
use_ema: True
ema_decay: 0.9999
ema_decay_type: "exponential"
act: silu
find_unused_parameters: True
depth_mult: 1.0 # no use in YOLOv7
width_mult: 1.0
arch: 'L'
use_aux: False
use_implicit: True
YOLOv7:
backbone: ELANNet
neck: ELANFPN
yolo_head: YOLOv7uHead # Anchor free, like PPYOLOE/YOLOv6/YOLOv8 Head
post_process: ~
ELANNet:
return_idx: [2, 3, 4]
depthwise: false
ELANFPN:
depthwise: false
for_u6: True #
YOLOv7uHead:
fpn_strides: [8, 16, 32]
loss_weight: {class: 0.5, iou: 7.5, dfl: 1.5}
assigner:
name: TaskAlignedAssigner
topk: 10
alpha: 0.5
beta: 6.0
nms:
name: MultiClassNMS
nms_top_k: 3000
keep_top_k: 300
score_threshold: 0.001
nms_threshold: 0.7
configs/yolov7u/_base_/yolov7u_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 290 # last 10 epochs close mosaic, totally 300 epochs as default
worker_num: 4
TrainReader:
sample_transforms:
- DecodeNormResize: {target_size: *input_size, mosaic: True}
- MosaicPerspective: {mosaic_prob: 1.0, target_size: *input_size, scale: 0.9, mixup_prob: 0.1, copy_paste_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: 8
shuffle: True
drop_last: False
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, 640, 640]
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/yolov7u/yolov7u_l_300e_coco.yml 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_300e.yml',
'_base_/yolov7u_elannet.yml',
'_base_/yolov7u_reader.yml',
]
log_iter: 100
snapshot_epoch: 10
weights: output/yolov7u_l_300e_coco/model_final
arch: L
TrainReader:
batch_size: 8 #16 # default 8 gpus, total bs = 128
configs/yolov8/0 0 → 100644
View file @ 522a602f
_BASE_: [
'../datasets/coco_detection.yml',
'../runtime.yml',
'_base_/optimizer_500e_high.yml',
'_base_/yolov8p6_cspdarknet.yml',
'_base_/yolov8p6_reader.yml',
]
depth_mult: 1.0 # not 1.33 as in YOLOv5-P6
width_mult: 1.25
log_iter: 10
snapshot_epoch: 10
weights: output/yolov8p6_x_500e_coco/model_final
YOLOv8CSPDarkNet:
last_stage_ch: 512 # The actual channel is int(512 * width_mult), not int(1024 * width_mult) as in YOLOv5-P6
last2_stage_ch: 512 # The actual channel is int(512 * width_mult), not int(768 * width_mult) as in YOLOv5-P6
TrainReader:
batch_size: 4 # default 8 gpus, total bs = 128
configs/yolov8/README.md 0 → 100644
View file @ 522a602f
# YOLOv8
## 内容
- [模型库](#模型库)
- [使用教程](#使用教程)
- [FastDeploy多硬件快速部署](#FastDeploy多硬件快速部署)
- [引用](#引用)
## 模型库
### 基础检测模型
| 网络网络 | 输入尺寸 | 图片数/GPU | 学习率策略 | TRT-FP16-Latency(ms) | mAP<sup>val<br>0.5:0.95 | mAP<sup>val<br>0.5 | Params(M) | FLOPs(G) | 下载链接 | 配置文件 |
| :------------- | :------- | :-------: | :------: | :------------: | :---------------------: | :----------------: |:---------: | :------: |:---------------: |:-----: |
| *YOLOv8-n | 640 | 16 | 500e | 1.8 | 37.3 | 53.0 | 3.16 | 8.7 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov8_n_500e_coco.pdparams) | [配置文件](./yolov8_n_500e_coco.yml) |
| *YOLOv8-s | 640 | 16 | 500e | 3.4 | 44.9 | 61.8 | 11.17 | 28.6 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov8_s_500e_coco.pdparams) | [配置文件](./yolov8_s_500e_coco.yml) |
| *YOLOv8-m | 640 | 16 | 500e | 6.5 | 50.2 | 67.3 | 25.90 | 78.9 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov8_m_500e_coco.pdparams) | [配置文件](./yolov8_m_500e_coco.yml) |
| *YOLOv8-l | 640 | 16 | 500e | 10.0 | 52.8 | 69.6 | 43.69 | 165.2 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov8_l_500e_coco.pdparams) | [配置文件](./yolov8_l_500e_coco.yml) |
| *YOLOv8-x | 640 | 16 | 500e | 15.1 | 53.8 | 70.6 | 68.23 | 257.8 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov8_x_500e_coco.pdparams) | [配置文件](./yolov8_x_500e_coco.yml) |
### Open Images v7 大规模600类检测模型
| 网络网络 | 输入尺寸 | 图片数/GPU | 学习率策略 | TRT-FP16-Latency(ms) | mAP<sup>val<br>0.5:0.95 | mAP<sup>val<br>0.5 | Params(M) | FLOPs(G) | 下载链接 | 配置文件 |
| :------------- | :------- | :-------: | :------: | :------------: | :---------------------: | :----------------: |:---------: | :------: |:---------------: |:-----: |
| *YOLOv8-n | 640 | 16 | 100e | 1.8 | - | - | 3.16 | 8.7 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov8_n_100e_oiv7.pdparams) | [配置文件](openimagev7/yolov8_n_100e_oiv7.yml) |
| *YOLOv8-s | 640 | 16 | 100e | 3.4 | - | - | 11.17 | 28.6 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov8_s_100e_oiv7.pdparams) | [配置文件](openimagev7/yolov8_s_100e_oiv7.yml) |
| *YOLOv8-m | 640 | 16 | 100e | 6.5 | - | - | 25.90 | 78.9 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov8_m_100e_oiv7.pdparams) | [配置文件](openimagev7/yolov8_m_100e_oiv7.yml) |
| *YOLOv8-l | 640 | 16 | 100e | 10.0 | - | - | 43.69 | 165.2 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov8_l_100e_oiv7.pdparams) | [配置文件](openimagev7/yolov8_l_100e_oiv7.yml) |
| *YOLOv8-x | 640 | 16 | 100e | 15.1 | - | - | 68.23 | 257.8 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov8_x_100e_oiv7.pdparams) | [配置文件](openimagev7/yolov8_x_100e_oiv7.yml) |
### 实例分割模型
| 网络网络 | 输入尺寸 | 图片数/GPU | 学习率策略 | TRT-FP16-Latency(ms) | box AP | mask AP | Params(M) | FLOPs(G) | 下载链接 | 配置文件 |
| :------------- | :------- | :-------: | :------: | :------------: | :---------------------: | :----------------: |:---------: | :------: |:---------------: |:-----: |
| *YOLOv8-n | 640 | 16 | 500e | - | 36.6 | - | 3.4 | 12.6 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov8_seg_n_500e_coco.pdparams) | [配置文件](../yolov8_seg/yolov8_seg_n_500e_coco.yml) |
| *YOLOv8-s | 640 | 16 | 500e | - | 44.6 | - | 11.8 | 42.6 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov8_seg_s_500e_coco.pdparams) | [配置文件](../yolov8_seg/yolov8_seg_s_500e_coco.yml) |
| *YOLOv8-m | 640 | 16 | 500e | - | 49.7 | - | 27.3 | 110.2 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov8_seg_m_500e_coco.pdparams) | [配置文件](../yolov8_seg/yolov8_seg_m_500e_coco.yml) |
| *YOLOv8-l | 640 | 16 | 500e | - | 52.1 | - | 46.0 | 220.5 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov8_seg_l_500e_coco.pdparams) | [配置文件](../yolov8_seg/yolov8_seg_l_500e_coco.yml) |
| *YOLOv8-x | 640 | 16 | 500e | - | 53.4 | - | 71.8 | 344.1 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov8_seg_x_500e_coco.pdparams) | [配置文件](../yolov8_seg/yolov8_seg_x_500e_coco.yml) |
### P6大尺度模型
| 网络网络 | 输入尺寸 | 图片数/GPU | 学习率策略 | TRT-FP16-Latency(ms) | mAP<sup>val<br>0.5:0.95 | mAP<sup>val<br>0.5 | Params(M) | FLOPs(G) | 下载链接 | 配置文件 |
| :------------- | :------- | :-------: | :------: | :------------: | :---------------------: | :----------------: |:---------: | :------: |:---------------: |:-----: |
| *YOLOv8-P6-x | 1280 | 16 | 500e | 55.0 | - | - | 97.42 | 522.93 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov8p6_x_500e_coco.pdparams) | [配置文件](./yolov8p6_x_500e_coco.yml) |
**注意:**
- YOLOv8模型mAP为部署权重在COCO val2017上的`mAP(IoU=0.5:0.95)`结果，且评估未使用`multi_label`等trick；
- YOLOv8模型训练使用COCO train2017作为训练集，Box AP为在COCO val2017上的`mAP(IoU=0.5:0.95)`结果；
- YOLOv8模型训练过程中默认使用8 GPUs进行混合精度训练，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；
- TRT-FP16-Latency(ms)模型推理耗时为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**。
- 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(带nms) | 导出后的权重(exclude_nms)| ONNX(exclude_post_process) |
| :-------- | :----: | :---------------: | :--------------------: | :-------------------------: |
| YOLOv8-n | 640 | [(w_nms)](https://paddledet.bj.bcebos.com/deploy/paddleyolo/yolov8/yolov8_n_500e_coco_w_nms.zip) | [(wo_nms)](https://paddledet.bj.bcebos.com/deploy/paddleyolo/yolov8/yolov8_n_500e_coco_wo_nms.zip) | [(onnx)](https://paddledet.bj.bcebos.com/deploy/paddleyolo/yolov8/yolov8_n_500e_coco.onnx) |
| YOLOv8-s | 640 | [(w_nms)](https://paddledet.bj.bcebos.com/deploy/paddleyolo/yolov8/yolov8_s_500e_coco_w_nms.zip) | [(wo_nms)](https://paddledet.bj.bcebos.com/deploy/paddleyolo/yolov8/yolov8_s_500e_coco_wo_nms.zip) | [(onnx)](https://paddledet.bj.bcebos.com/deploy/paddleyolo/yolov8/yolov8_s_500e_coco.onnx) |
| YOLOv8-m | 640 | [(w_nms)](https://paddledet.bj.bcebos.com/deploy/paddleyolo/yolov8/yolov8_m_500e_coco_w_nms.zip) | [(wo_nms)](https://paddledet.bj.bcebos.com/deploy/paddleyolo/yolov8/yolov8_m_500e_coco_wo_nms.zip) | [(onnx)](https://paddledet.bj.bcebos.com/deploy/paddleyolo/yolov8/yolov8_m_500e_coco.onnx) |
| YOLOv8-l | 640 | [(w_nms)](https://paddledet.bj.bcebos.com/deploy/paddleyolo/yolov8/yolov8_l_500e_coco_w_nms.zip) | [(wo_nms)](https://paddledet.bj.bcebos.com/deploy/paddleyolo/yolov8/yolov8_l_500e_coco_wo_nms.zip) | [(onnx)](https://paddledet.bj.bcebos.com/deploy/paddleyolo/yolov8/yolov8_l_500e_coco.onnx) |
| YOLOv8-x | 640 | [(w_nms)](https://paddledet.bj.bcebos.com/deploy/paddleyolo/yolov8/yolov8_x_500e_coco_w_nms.zip) | [(wo_nms)](https://paddledet.bj.bcebos.com/deploy/paddleyolo/yolov8/yolov8_x_500e_coco_wo_nms.zip) | [(onnx)](https://paddledet.bj.bcebos.com/deploy/paddleyolo/yolov8/yolov8_x_500e_coco.onnx) |
**注意:**
- 带nms的导出权重为普通导出方式，加trt表示用于trt加速，对NMS和silu激活函数提速明显。运行命令为：
```CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} trt=True```
- `exclude_nms`导出的权重表示去除NMS导出，返回2个Tensor，是缩放回原图后的坐标和分类置信度。运行命令为：
```CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} exclude_nms=True trt=True```
- `exclude_post_process`导出表示去除后处理导出，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和分类置信度。运行命令为：
```CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} exclude_post_process=True trt=True ```
## 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=yolov8 # 可修改，如 ppyoloe
job_name=yolov8_s_500e_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YOLOv8
```bash
python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 tools/train.py -c configs/yolov8/yolov8_s_500e_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/yolov8/yolov8_s_500e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov8_s_500e_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/yolov8/yolov8_s_500e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov8_s_500e_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/yolov8/yolov8_s_500e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov8_s_500e_coco.pdparams --infer_dir=demo
```
### 4.导出模型
YOLOv8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/yolov8/yolov8_s_500e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov8_s_500e_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/yolov8/yolov8_s_500e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov8_s_500e_coco.pdparams trt=True
```
如果你想将YOLOv8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/yolov8/yolov8_s_500e_coco.yml --output_dir=output_inference -o weights=https://paddledet.bj.bcebos.com/models/yolov8_s_500e_coco.pdparams
# 安装paddle2onnx
pip install paddle2onnx
# 转换成onnx格式
paddle2onnx --model_dir output_inference/yolov8_s_500e_coco --model_filename model.pdmodel --params_filename model.pdiparams --opset_version 11 --save_file yolov8_s_500e_coco.onnx
```
**注意：** ONNX模型目前只支持batch_size=1
## FastDeploy多硬件快速部署
FastDeploy是飞桨推出的统一部署工具，支持云边端部署。目前在YOLO系列支持的部署能力如下所示。具体部署示例，可以前往[FastDeploy仓库](https://github.com/PaddlePaddle/FastDeploy)使用。
| | [YOLOv5](https://github.com/PaddlePaddle/FastDeploy/tree/develop/examples/vision/detection/paddledetection) | [YOLOv6](https://github.com/PaddlePaddle/FastDeploy/tree/develop/examples/vision/detection/paddledetection) | [YOLOv7](https://github.com/PaddlePaddle/FastDeploy/tree/develop/examples/vision/detection/paddledetection) | [YOLOv8](https://github.com/PaddlePaddle/FastDeploy/tree/develop/examples/vision/detection/paddledetection) | [PP-YOLOE+](https://github.com/PaddlePaddle/FastDeploy/tree/develop/examples/vision/detection/paddledetection) | 部署特色 |
| ------------------------------------------------------------------------------------------------------------------------------ | ----------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------- | ------------------------------------- |
| [Intel CPU](https://github.com/PaddlePaddle/FastDeploy/blob/develop/docs/cn/build_and_install/download_prebuilt_libraries.md) | 支持 | 支持 | 支持 | 支持 | 支持 | 集成PaddleSlim一键压缩压缩，实现极致性能 |
| [NVIDIA GPU](https://github.com/PaddlePaddle/FastDeploy/blob/develop/docs/cn/build_and_install/download_prebuilt_libraries.md) | 支持 | 支持 | 支持 | 支持 | 支持 | 集成PaddleSlim一键压缩工具、CUDA预处理加速，实现极致性能 |
| [飞腾 CPU]() | 支持 | 支持 | 支持 | 支持 | 支持 | X86 CPU与ARM CPU无缝切换 |
| [昆仑芯 R200*](https://github.com/PaddlePaddle/FastDeploy/blob/develop/docs/cn/build_and_install/kunlunxin.md) | 支持 | 支持 | 支持 | 支持 | 支持 | 无缝部署Paddle模型 |
| [昇腾310*](https://github.com/PaddlePaddle/FastDeploy/blob/develop/docs/cn/build_and_install/kunlunxin.md) | 支持 | 即将支持 | 即将支持 | 即将支持 | 支持 | 无缝部署Paddle模型 |
| [算能SC7-FP300*](https://github.com/PaddlePaddle/FastDeploy/blob/develop/docs/cn/build_and_install/sophgo.md) | 支持 | 支持 | 支持 | 支持 | 支持 | 充分发挥硬件工具链特性，实现模型快速部署 |
| [Jetson](https://github.com/PaddlePaddle/FastDeploy/blob/develop/docs/cn/build_and_install/jetson.md) | 支持 | 支持 | 支持 | 支持 | 支持 | 集成PaddleSlim一键压缩工具、CUDA预处理加速，实现极致性能 |
| [ARM CPU](https://github.com/PaddlePaddle/FastDeploy/blob/develop/docs/cn/build_and_install/download_prebuilt_libraries.md) | 支持 | 支持 | 支持 | 支持 | 支持 | 集成PaddleSlim一键压缩工具、预处理加速库FlyCV，实现极致性能 |
| [RK3588*](https://github.com/PaddlePaddle/FastDeploy/blob/develop/docs/cn/build_and_install/rknpu2.md) | 支持 | 支持 | 支持 | 支持 | 支持 | 充分发挥硬件工具链特性，实现模型快速部署 |
| [RV1126*](https://github.com/PaddlePaddle/FastDeploy/blob/develop/docs/cn/build_and_install/rv1126.md) | 支持 | 暂不支持 | 暂不支持 | 暂不支持 | 支持 | 联合全量化实现模型端到端的优化 |
| [服务化部署](https://github.com/PaddlePaddle/FastDeploy/tree/develop/serving) | 支持 | 暂不支持 | 暂不支持 | 暂不支持 | 支持 | 实现企业级高并发需求 |
| [视频流部署](https://github.com/PaddlePaddle/FastDeploy/tree/develop/streamer) | 暂不支持 | 暂不支持 | 暂不支持 | 暂不支持 | 支持 | 调用硬解码核，实现数据零拷贝，充分利用硬件资源 |
备注：
*表示：FastDeploy目前在该型号硬件上测试。通常同类型硬件上使用的是相同的软件栈，该部署能力可以延伸到同软件架栈的硬件。譬如RK3588与RK3566、RK3568相同的软件栈。
「硬件列-纵轴」链接到部署预编译包安装或部署示例，「横轴」跳转到具体部署示例。
## 引用
```
```
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