yaclty/FastestDet
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

update

Browse Source
This commit is contained in:
duoduo 2022-07-07 15:54:48 +08:00
parent 48911cceae
commit a2883e2bb7
15 changed files with 35 additions and 37 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

52
README.md
View File

@ -1,12 +1,11 @@
！！！！因为ncnn onnx2ncnn的bug导致crop算子计算有问题，故模型的指标需重新测试:https://github.com/Tencent/ncnn/pull/3999
！！！！因为ncnn onnx2ncnn的bug导致crop算子计算有问题，修复后模型的指标作了调整修正:https://github.com/Tencent/ncnn/pull/3999
！！！！Because the bug of ncnn onnx2ncnn causes problems in the calculation of the crop operator, the indicators of the model need to be retested :https://github.com/Tencent/ncnn/pull/3999
# :zap:FastestDet:zap:[![DOI](https://zenodo.org/badge/508635170.svg)](https://zenodo.org/badge/latestdoi/508635170)
![image](https://github.com/dog-qiuqiu/FastestDet/blob/main/data/data.png)
* ***Faster! Stronger! Simpler!***
* ***It has better single core reasoning performance and simpler feature map post-processing than Yolo-fastest***
* ***In the ARM CPU of RK3568, the single core reasoning performance is 50% higher than Yolo-fastest***
* ***The coco evaluation index increased by 3.8% compared with the map0.5 of Yolo-fastest***
* ***It has better performance and simpler feature map post-processing than Yolo-fastest***
* ***The performance is 10% higher than Yolo-fastest***
* ***The coco evaluation index increased by 0.9% compared with the map0.5 of Yolo-fastestv2***
* ***算法介绍：https://zhuanlan.zhihu.com/p/536500269 交流qq群:1062122604***
# Evaluating indicator/Benchmark
Network|mAPval 0.5|mAPval 0.5:0.95|Resolution|Run Time(4xCore)|Run Time(1xCore)|Params(M)
@ -17,7 +16,7 @@ Network|mAPval 0.5|mAPval 0.5:0.95|Resolution|Run Time(4xCore)|Run Time(1xCore)|
[nanodet_m](https://github.com/RangiLyu/nanodet)|-|20.6%|320X320|49.24ms|160.35ms|0.95M
[yolo-fastestv1.1](https://github.com/dog-qiuqiu/Yolo-Fastest/tree/master/ModelZoo/yolo-fastest-1.1_coco)|24.40%|-|320X320|26.60ms|75.74ms|0.35M
[yolo-fastestv2](https://github.com/dog-qiuqiu/Yolo-FastestV2/tree/main/modelzoo)|24.10%|-|352X352|23.8ms|68.9ms|0.25M
FastestDet|27.8%|14.0%|512X512|21.51ms|34.62ms|0.25M
FastestDet|25.0%|12.3%|352X352|23.51ms|70.62ms|0.24M
* ***Test platform Radxa Rock3A RK3568 ARM Cortex-A55 CPU，Based on [NCNN](https://github.com/Tencent/ncnn)***
* ***CPU lock frequency 2.0GHz***
# Improvement
@ -28,10 +27,9 @@ FastestDet|27.8%|14.0%|512X512|21.51ms|34.62ms|0.25M
# Multi-platform benchmark
Equipment|Computing backend|System|Framework|Run time(Single core)|Run time(Multi core)
:---:|:---:|:---:|:---:|:---:|:---:
Radxa rock3a|RK3568(arm-cpu)|Linux(aarch64)|ncnn|34.62ms|21.51ms
Qualcomm|Snapdragon 835(arm-cpu)|Android(aarch64)|ncnn|28.34ms|16.24ms
AMD|R5-5600(X86-cpu)|Linux(amd64)|ncnn|2.16ms|1.73ms
Intel|i7-8700(X86-cpu)|Linux(amd64)|ncnn|5.21ms|4.73ms
Radxa rock3a|RK3568(arm-cpu)|Linux(aarch64)|ncnn|70.62ms|23.51ms
Qualcomm|Snapdragon 835(arm-cpu)|Android(aarch64)|ncnn|32.34ms|16.24ms
Intel|i7-8700(X86-cpu)|Linux(amd64)|ncnn|4.51ms|4.33ms
# How to use
## Dependent installation
* PiP(Note pytorch CUDA version selection)
@ -41,7 +39,7 @@ Intel|i7-8700(X86-cpu)|Linux(amd64)|ncnn|5.21ms|4.73ms
## Test
* Picture test
```
python3 test.py --yaml configs/coco.yaml --weight weights/weight_AP05\:0.278_280-epoch.pth --img data/3.jpg
python3 test.py --yaml configs/coco.yaml --weight weights/coco_ap05_0.250_280epoch.pth --img data/3.jpg
```
<div align=center>
<img src="https://github.com/dog-qiuqiu/FastestDet/blob/main/result.png"> />
@ -128,8 +126,8 @@ DATASET:
NAMES: "dataset/coco128/coco.names" # .names category label file
MODEL:
NC: 80 # Number of detection categories
INPUT_WIDTH: 512 # The width of the model input image
INPUT_HEIGHT: 512 # The height of the model input image
INPUT_WIDTH: 352 # The width of the model input image
INPUT_HEIGHT: 352 # The height of the model input image
TRAIN:
LR: 0.001 # Train learn rate
THRESH: 0.25 # ？？？？
@ -149,7 +147,7 @@ TRAIN:
### Evaluation
* Calculate map evaluation
```
python3 eval.py --yaml configs/coco.yaml --weight weights/weight_AP05\:0.278_280-epoch.pth
python3 eval.py --yaml configs/coco.yaml --weight weights/coco_ap05_0.250_280epoch.pth
```
* COCO2017 evaluation
```
@ -162,25 +160,25 @@ TRAIN:
DONE (t=30.85s).
Accumulating evaluation results...
DONE (t=4.97s).
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.140
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.278
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.128
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.018
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.103
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.232
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.157
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.225
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.231
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.032
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.201
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.359
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.123
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.250
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.109
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.017
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.115
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.238
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.139
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.199
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.205
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.035
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.218
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.374
```
# Deploy
## Export onnx
* You can export .onnx by adding the --onnx option when executing test.py
```
python3 test.py --yaml configs/coco.yaml --weight weights/weight_AP05\:0.278_280-epoch.pth --img data/3.jpg --onnx
python3 test.py --yaml configs/coco.yaml --weight weights/coco_ap05_0.250_280epoch.pth --img data/3.jpg --onnx
```
## onnx-runtime
* You can learn about the pre and post-processing methods of FastestDet in this Sample

14
configs/coco.yaml
View File

@ -4,15 +4,15 @@ DATASET:
NAMES: "configs/coco.names"
MODEL:
NC: 80
INPUT_WIDTH: 512
INPUT_HEIGHT: 512
INPUT_WIDTH: 352
INPUT_HEIGHT: 352
TRAIN:
LR: 0.001
THRESH: 0.25
WARMUP: true
BATCH_SIZE: 64
END_EPOCH: 350
BATCH_SIZE: 96
END_EPOCH: 300
MILESTIONES:
- 150
- 250
- 300
- 100
- 200
- 250

BIN
data/0_result.png
View File

Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 630 KiB

After

Width:  |  Height:  |  Size: 633 KiB

BIN
data/1_result.png
View File

Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 585 KiB

After

Width:  |  Height:  |  Size: 586 KiB

BIN
data/2_result.png
View File

Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 577 KiB

After

Width:  |  Height:  |  Size: 578 KiB

BIN
data/3_result.png
View File

Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 290 KiB

After

Width:  |  Height:  |  Size: 290 KiB

BIN
data/4_result.png
View File

Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 525 KiB

After

Width:  |  Height:  |  Size: 525 KiB

BIN
example/onnx-runtime/FastestDet.onnx
View File

Binary file not shown.

BIN
example/onnx-runtime/result.jpg
View File

Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 161 KiB

After

Width:  |  Height:  |  Size: 157 KiB

2
example/onnx-runtime/runtime.py
View File

@ -116,7 +116,7 @@ if __name__ == '__main__':
# 读取图片
img = cv2.imread("1.jpg")
# 模型输入的宽高
input_width, input_height = 512, 512
input_width, input_height = 352, 352
# 加载模型
session = onnxruntime.InferenceSession('FastestDet.onnx')
# 目标检测

BIN
result.png
View File

Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 290 KiB

After

Width:  |  Height:  |  Size: 290 KiB

2
test.py
View File

@ -13,7 +13,7 @@ if __name__ == '__main__':
parser.add_argument('--yaml', type=str, default="", help='.yaml config')
parser.add_argument('--weight', type=str, default=None, help='.weight config')
parser.add_argument('--img', type=str, default='', help='The path of test image')
parser.add_argument('--thresh', type=float, default=0.8, help='The path of test image')
parser.add_argument('--thresh', type=float, default=0.65, help='The path of test image')
parser.add_argument('--onnx', action="store_true", default=False, help='Export onnx file')
parser.add_argument('--cpu', action="store_true", default=False, help='Run on cpu')

2
utils/tool.py
View File

@ -58,7 +58,7 @@ class EMA():
self.backup = {}
# 后处理(归一化后的坐标)
def handel_preds(preds, device, conf_thresh=0.25, nms_thresh=0.35):
def handel_preds(preds, device, conf_thresh=0.25, nms_thresh=0.45):
total_bboxes, output_bboxes = [], []
# 将特征图转换为检测框的坐标
N, C, H, W = preds.shape

BIN
weights/coco_ap05_0.250_280epoch.pth Normal file
View File

Binary file not shown.

BIN
weights/weight_AP05:0.278_280-epoch.pth
View File

Binary file not shown.