🍅🍅🍅YOLOv5-Lite: lighter, faster and easier to deploy. Evolved from yolov5 and the size of model is only 1.7M (int8) and 3.3M (fp16). It can reach 10+ FPS on the Raspberry Pi 4B when the input size is 320×320~

Last update: Jan 5, 2023

Overview

YOLOv5-Lite：lighter, faster and easier to deploy

Perform a series of ablation experiments on yolov5 to make it lighter (smaller Flops, lower memory, and fewer parameters) and faster (add shuffle channel, yolov5 head for channel reduce. It can infer at least 10+ FPS On the Raspberry Pi 4B when input the frame with 320×320) and is easier to deploy (removing the Focus layer and four slice operations, reducing the model quantization accuracy to an acceptable range).

Comparison of ablation experiment results

ID	Model	Input_size	Flops	Params	Size（M）	[email protected]	[email protected]:0.95
001	yolo-fastest	320×320	0.25G	0.35M	1.4	24.4	-
002	nanodet-m	320×320	0.72G	0.95M	1.8	-	20.6
003	yolo-fastest-xl	320×320	0.72G	0.92M	3.5	34.3	-
004	yolov5-lite	320×320	1.43G	1.62M	3.3	36.2	20.8
005	yolov3-tiny	416×416	6.96G	6.06M	23.0	33.1	16.6
006	yolov4-tiny	416×416	5.62G	8.86M	33.7	40.2	21.7
007	nanodet-m	416×416	1.2G	0.95M	1.8	-	23.5
008	yolov5-lite	416×416	2.42G	1.62M	3.3	41.3	24.4
009	yolov5-lite	640×640	2.42G	1.62M	3.3	45.7	27.1
010	yolov5s	640×640	17.0G	7.3M	14.2	55.4	36.7

Comparison on different platforms

Equipment	Computing backend	System	Framework	Input	Speed{our}	Speed{yolov5s}
Inter	@i5-10210U	window(x86)	640×640	torch-cpu	112ms	179ms
Nvidia	@RTX 2080Ti	Linux(x86)	640×640	torch-gpu	11ms	13ms
Raspberrypi 4B	@ARM Cortex-A72	Linux(arm64)	320×320	ncnn	97ms	371ms

Detection effect

Pytorch{640×640}：

NCNN{FP16}@{640×640}:

NCNN{Int8}@{640×640}:

Base on YOLOv5

10FPS can be used with yolov5 on the Raspberry Pi with only 0.1T computing power

Excluding the first three warm-ups, the device temperature is stable above 45°, the forward reasoning framework is ncnn, and the two benchmark comparisons are recorded

# 第四次
pi@raspberrypi:~/Downloads/ncnn/build/benchmark $ ./benchncnn 8 4 0
loop_count = 8
num_threads = 4
powersave = 0
gpu_device = -1
cooling_down = 1
         yolov5-lite  min =   90.86  max =   93.53  avg =   91.56
    yolov5-lite-int8  min =   83.15  max =   84.17  avg =   83.65
     yolov5-lite-416  min =  154.51  max =  155.59  avg =  155.09
         yolov4-tiny  min =  298.94  max =  302.47  avg =  300.69
           nanodet_m  min =   86.19  max =  142.79  avg =   99.61
          squeezenet  min =   59.89  max =   60.75  avg =   60.41
     squeezenet_int8  min =   50.26  max =   51.31  avg =   50.75
           mobilenet  min =   73.52  max =   74.75  avg =   74.05
      mobilenet_int8  min =   40.48  max =   40.73  avg =   40.63
        mobilenet_v2  min =   72.87  max =   73.95  avg =   73.31
        mobilenet_v3  min =   57.90  max =   58.74  avg =   58.34
          shufflenet  min =   40.67  max =   41.53  avg =   41.15
       shufflenet_v2  min =   30.52  max =   31.29  avg =   30.88
             mnasnet  min =   62.37  max =   62.76  avg =   62.56
     proxylessnasnet  min =   62.83  max =   64.70  avg =   63.90
     efficientnet_b0  min =   94.83  max =   95.86  avg =   95.35
   efficientnetv2_b0  min =  103.83  max =  105.30  avg =  104.74
        regnety_400m  min =   76.88  max =   78.28  avg =   77.46
           blazeface  min =   13.99  max =   21.03  avg =   15.37
           googlenet  min =  144.73  max =  145.86  avg =  145.19
      googlenet_int8  min =  123.08  max =  124.83  avg =  123.96
            resnet18  min =  181.74  max =  183.07  avg =  182.37
       resnet18_int8  min =  103.28  max =  105.02  avg =  104.17
             alexnet  min =  162.79  max =  164.04  avg =  163.29
               vgg16  min =  867.76  max =  911.79  avg =  889.88
          vgg16_int8  min =  466.74  max =  469.51  avg =  468.15
            resnet50  min =  333.28  max =  338.97  avg =  335.71
       resnet50_int8  min =  239.71  max =  243.73  avg =  242.54
      squeezenet_ssd  min =  179.55  max =  181.33  avg =  180.74
 squeezenet_ssd_int8  min =  131.71  max =  133.34  avg =  132.54
       mobilenet_ssd  min =  151.74  max =  152.67  avg =  152.32
  mobilenet_ssd_int8  min =   85.51  max =   86.19  avg =   85.77
      mobilenet_yolo  min =  327.67  max =  332.85  avg =  330.36
  mobilenetv2_yolov3  min =  221.17  max =  224.84  avg =  222.60

# 第八次
pi@raspberrypi:~/Downloads/ncnn/build/benchmark $ ./benchncnn 8 4 0
loop_count = 8
num_threads = 4
powersave = 0
gpu_device = -1
cooling_down = 1
           nanodet_m  min =   81.15  max =   81.71  avg =   81.33
       nanodet_m-416  min =  143.89  max =  145.06  avg =  144.67
         yolov5-lite  min =   84.30  max =   86.34  avg =   85.79
    yolov5-lite-int8  min =   80.98  max =   82.80  avg =   81.25
     yolov5-lite-416  min =  142.75  max =  146.10  avg =  144.34
         yolov4-tiny  min =  276.09  max =  289.83  avg =  285.99
          squeezenet  min =   59.37  max =   61.19  avg =   60.35
     squeezenet_int8  min =   49.30  max =   49.66  avg =   49.43
           mobilenet  min =   72.40  max =   74.13  avg =   73.37
      mobilenet_int8  min =   39.92  max =   40.23  avg =   40.07
        mobilenet_v2  min =   71.57  max =   73.07  avg =   72.29
        mobilenet_v3  min =   54.75  max =   56.00  avg =   55.40
          shufflenet  min =   40.07  max =   41.13  avg =   40.58
       shufflenet_v2  min =   29.39  max =   30.25  avg =   29.86
             mnasnet  min =   59.54  max =   60.18  avg =   59.96
     proxylessnasnet  min =   61.06  max =   62.63  avg =   61.75
     efficientnet_b0  min =   91.86  max =   95.01  avg =   92.84
   efficientnetv2_b0  min =  101.03  max =  102.61  avg =  101.71
        regnety_400m  min =   76.75  max =   78.58  avg =   77.60
           blazeface  min =   13.18  max =   14.67  avg =   13.79
           googlenet  min =  136.56  max =  138.05  avg =  137.14
      googlenet_int8  min =  118.30  max =  120.17  avg =  119.23
            resnet18  min =  164.78  max =  166.80  avg =  165.70
       resnet18_int8  min =   98.58  max =   99.23  avg =   98.96
             alexnet  min =  155.06  max =  156.28  avg =  155.56
               vgg16  min =  817.64  max =  832.21  avg =  827.37
          vgg16_int8  min =  457.04  max =  465.19  avg =  460.64
            resnet50  min =  318.57  max =  323.19  avg =  320.06
       resnet50_int8  min =  237.46  max =  238.73  avg =  238.06
      squeezenet_ssd  min =  171.61  max =  173.21  avg =  172.10
 squeezenet_ssd_int8  min =  128.01  max =  129.58  avg =  128.84
       mobilenet_ssd  min =  145.60  max =  149.44  avg =  147.39
  mobilenet_ssd_int8  min =   82.86  max =   83.59  avg =   83.22
      mobilenet_yolo  min =  311.95  max =  374.33  avg =  330.15
  mobilenetv2_yolov3  min =  211.89  max =  286.28  avg =  228.01

NCNN_Android_demo

This is a Redmi phone, the processor is Snapdragon 730G, and yolov5-lite is used for detection. The performance is as follows:

This is the quantized int8 model:

Outdoor scene example:

More detailed explanation

Detailed model link: https://zhuanlan.zhihu.com/p/400545131

NCNN deployment and int8 quantization：https://zhuanlan.zhihu.com/p/400975662

Reference

https://github.com/Tencent/ncnn

https://github.com/ultralytics/yolov5

https://github.com/megvii-model/ShuffleNet-Series

Comments

在COCO上直接训练v5lite-s 416x416，未修改任何参数，map仅35.2

直接用原始参数在COCO上训练v5lite-s，输入416x416，测试的结果如下： 模型测试命令：python test.py --device 0 --conf-thres 0.1 --iou-thres 0.5

Class Images Labels P R [email protected] [email protected]:.95: 100%|█| 79/79 [00:45<00:00 all 5000 36335 0.537 0.363 0.352 0.203

使用博主提供的模型v5lite-s，输入416x416，测试的结果如下： 模型测试命令：python test.py --device 0 --conf-thres 0.1 --iou-thres 0.5

Class Images Labels P R [email protected] [email protected]:.95: 100%|█| 79/79 [00:48<00:00 all 5000 36335 0.542 0.388 0.373 0.225

mAP相差2个点，请问这是什么原因导致的呢？期待大佬的回复！谢谢。
documentation

opened by Broad-sky 12
speed problem
使用原始yolov5s6 640输入，速度为12ms，使用下面repvgg_block 640输入为13ms GPU v100

YOLOv5 🚀 by Ultralytics, GPL-3.0 license

Parameters

nc: 80 # number of classes depth_multiple: 0.33 # model depth multiple width_multiple: 0.50 # layer channel multiple anchors:

[19,27, 44,40, 38,94] # P3/8

[96,68, 86,152, 180,137] # P4/16

[140,301, 303,264, 238,542] # P5/32

[436,615, 739,380, 925,792] # P6/64

YOLOv5 v6.0 backbone

backbone:

[from, number, module, args]

[[-1, 1, Conv, [32, 6, 2, 2]], # 0-P1/2 [-1, 1, Conv, [64, 3, 2]], # 1-P2/4 [-1, 1, C3, [64]], [-1, 1, RepVGGBlock, [128, 3, 2]], # 3-P3/8 [-1, 3, C3, [128]], [-1, 1, RepVGGBlock, [256, 3, 2]], # 5-P4/16 [-1, 3, C3, [256]], [-1, 1, RepVGGBlock, [512, 3, 2]], # 7-P5/32 [-1, 3, C3, [512]], [-1, 1, RepVGGBlock, [768, 3, 2]], # 9-P6/64 [-1, 3, C3, [768]], [-1, 1, SPPF, [768, 5]], # 11 ]

YOLOv5 v6.0 head

head: [[-1, 1, Conv, [512, 1, 1]], [-1, 1, nn.Upsample, [None, 2, 'nearest']], [[-1, 8], 1, Concat, [1]], # cat backbone P5 [-1, 3, C3, [512, False]], # 15

[-1, 1, Conv, [256, 1, 1]], [-1, 1, nn.Upsample, [None, 2, 'nearest']], [[-1, 6], 1, Concat, [1]], # cat backbone P4 [-1, 3, C3, [256, False]], # 19

[-1, 1, Conv, [128, 1, 1]], [-1, 1, nn.Upsample, [None, 2, 'nearest']], [[-1, 4], 1, Concat, [1]], # cat backbone P3 [-1, 3, C3, [128, False]], # 23 (P3/8-small)

[-1, 1, Conv, [128, 3, 2]], [[-1, 20], 1, Concat, [1]], # cat head P4 [-1, 3, C3, [256, False]], # 26 (P4/16-medium)

[-1, 1, Conv, [256, 3, 2]], [[-1, 16], 1, Concat, [1]], # cat head P5 [-1, 3, C3, [512, False]], # 29 (P5/32-large)

[-1, 1, Conv, [512, 3, 2]], [[-1, 12], 1, Concat, [1]], # cat head P6 [-1, 3, C3, [768, False]], # 32 (P6/64-xlarge)

[[23, 26, 29, 32], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5, P6) ]
opened by Hiwyl 10
关于使用c++部署的问题？

你好，大佬，就是我想问一下如果使用trt或tensorrt加速能够用c++的api吗，因为目前我的项目其他部分是用的c++，然后用tensorrt加速yolov5用的是c++，主要是我看您这个把pplcnet和yolov5s结合的工作在速度上有很大提升，我就是想用openvino的c++的api来推理可以吗？

opened by Hezhexi2002 10

使用openvino推理xml文件报错

我在项目根目录下python openvino/openvino.py -m v5lite-c.xml -i openvino/bike.jpg, 我事先运行了/opt/intel/openvino_2021/bin/setupvars.sh 报错：

Traceback (most recent call last):
  File "openvino/openvino.py", line 23, in <module>
    import ngraph
  File "/opt/intel/openvino_2021/python/python3.8/ngraph/__init__.py", line 16, in <module>
    from ngraph.helpers import function_from_cnn
  File "/opt/intel/openvino_2021/python/python3.8/ngraph/helpers.py", line 7, in <module>
    from openvino.inference_engine import IENetwork
  File "/home/fraunhofer/Software/YOLOv5-Lite/openvino/openvino.py", line 25, in <module>
    from openvino.inference_engine import IENetwork, IECore
ModuleNotFoundError: No module named 'openvino.inference_engine'; 'openvino' is not a package

opened by ghost 8

Attention imporved yolov5 performance

hi. i am interested in this project. maybe i can contribute for this good job.

| Model | map0.5:0.95 | map0.5:0.05 | Speed CPU b1 | Speed 2080ti b1 | Speed 2080ti b32 |params(M) | FLOPS | | ---------- | :-----------: | :-----------: | :-----------: | :-----------: | :-----------: | :-----------: | :-----------: | | 5s offical | 37.2 | 56.0 | 98 | 6.4(V100) | 0.9(V100) | 7.2 | 16.5 | | 5s | 37.2 | 56.8 | 47.7 | 7.3 | 1.0 | 7.2 | 16.5 | | 5sbackbone + C3GC | 37.7 | 57.3 | 63.6 | 8.4 | 1.4 | 7.5 | 16.8 | | 5sC3GC + FPN_conv | 40.5 | 59.6 | 92.7.0 | 12.0 | 1.5 | 11.0 | 22.7 | | 5m | 45.2 | 63.9 | 224 | 8.2(V100) | 1.7(V100) | 21.2 | 49.0 |

opened by 315386775 7
MNN performance much less than NCNN

Hi, has anybody met the problem that MNN has less performance than NCNN？ Like the confidence and mAP? I printed out the outputs of NCNN and MNN in terms of the same image, but their outputs differ a lot. The confidences of MNN are almost all lower than NCNN, and bboxs are worse as well. Another question is that my input size of an image is 320 * 240, but the model input size is 320 * 256. Is that reasonable?

opened by bl6g6 7
模型文件在master版本的yolo下面跑不起来

python3 train.py --data data/data.yaml --cfg models/v5Lite-g.yaml --batch-size 8

RuntimeError: Given groups=1, weight of size [3, 64, 2, 2], expected input[1, 32, 128, 128] to have 64 channels, but got 32 channels instead

opened by deep-practice 7
测试帧率问题

您好，复现shufflev2-yolov5代码与官方yolov5s代码对比，在同一实验环境下测试同一段视频为何测试速度反而是yolov5s更快一些（测试了两次）。

shufflev2-yolov5 | yolov5s ------------ | ------------- 96.353s | 92.874s 95.978s | 90.501s

opened by luckywangchenxi 7
shufflenetv2预训练的模型的问题

请问您有用shufflenetv2的预训练模型训练吗？我要用pytorch官网的'shufflenetv2_x1.0' 模型的时候出现层的shape对应不上的问题：左边是官方的，右边是你的 10 : stage2.0.branch1.2.weight torch.Size([58, 24, 1, 1]) __ model.1.branch1.1.running_var torch.Size([24]) 11 : stage2.0.branch1.3.weight torch.Size([58]) __ model.1.branch1.1.num_batches_tracked torch.Size([]) 12 : stage2.0.branch1.3.bias torch.Size([58]) __ model.1.branch1.2.weight torch.Size([60, 24, 1, 1]

你的模型会出现多出一层‘model.1.branch1.1.num_batches_tracked’ 而且shape是[ ]空的，网上说这是新版pytorch的问题但是主要问题是左边stage2.0.branch1.2 ([58, 24, 1, 1]) 和右边你的branch1.2.weight torch.Size([60, 24, 1, 1] 开始出现shape不匹配，接下去陆陆续续都开始shape不一样了

是您搭建网络出错了吗？

opened by chenweifu2008 7
v5Lite-g模型训练时map很低。

我正在进行该项目思路相同的工作，我认为将repvgg嵌入到yolov5中是一种很好的想法，根据v5Lite-g.yaml提供的模型结构，我使用nn.model重新构建了该结构，并将提供的权重转换到我重构的模型结构内，在我的测试中两个模型具有完全相同的检测结果，姑且我认为以上的工作，我是正确的。但是当使用我自己重构的结构来训练模型时，我发现得到的map(Terminal截图)会比该项目使用lite-g训练的map(日志截图)会低很多，即使我使用了完全相同的训练配置。 Yolov5-Lite项目基于v5 5.0版本开发，请问，相较u版的代码，该项目是否在lite-g模型测试时添加了其他额外的操作，或者额外的方法呢？

opened by Materx 6
unexpected result

Hello

Sorry if this is more a question that an issue, i'm not sure if i'm doing something wrong.

I trained a new model, using the v5lite-s weight with "person" class using coco + voc dataset, the 1k background images as negative samples and to have an input of 320. I did not change any hyperparameter and using ncnn framework

I get: mAP 05:0.95 of 0.498 mAP 05 of 0.795 Precision 0.84 Recall 0.68

I was expecting to have a better result than the provided model trained on all coco classes... but as you can see there is at least two persons not detected.

Maybe is a problem on the model conversion?

Thanks

Standard coco model

New coco_voc model

opened by natxopedreira 6

TypeError: argument of type 'int' is not iterable

in linux and windows,both are occured error

Namespace(agnostic_nms=False, augment=False, classes=None, conf_thres=0.45, device='', exist_ok=False, img_size=640, iou_thres=0.5, name='exp', nosave=False, project='runs/detect', save_conf=False, save_txt=False, source='0', update=False, view_img=False, weights='weights/v5lite-s.pt')
YOLOv5  v1.4-51-gca7ed7c torch 1.13.0+cpu CPU

Fusing layers...
Model Summary: 230 layers, 1640661 parameters, 0 gradients
1/1: 0... Traceback (most recent call last):
  File "detect.py", line 178, in <module>
    detect()
  File "detect.py", line 51, in detect
    dataset = LoadStreams(source, img_size=imgsz, stride=stride)
  File "D:\er_code\yolov5-lite\utils\datasets.py", line 279, in __init__
    if 'youtube.com/' in url or 'youtu.be/' in url:  # if source is YouTube video
TypeError: argument of type 'int' is not iterable

another question,is that possible for me to load the model by hub or something else?

opened by danyow-cheung 0

Axera's AX620A has supported the YOLOv5-Lite

repo

https://github.com/AXERA-TECH/ax-samples

source code

https://github.com/AXERA-TECH/ax-samples/blob/main/examples/ax_yolov5_lite_steps.cc

model

https://github.com/AXERA-TECH/ax-models/blob/main/ax620/v5Lite-g-sim-640.joint

result on AXera-Pi

https://github.com/AXERA-TECH/ax-samples/tree/main/examples#yolov5-lite

root@AXERA:~/samples# ./ax_yolov5_lite -i cengiz-sari-X4spr8Kuwxc-unsplash.jpg -m ./models/v5Lite-g-sim-640.joint
--------------------------------------
model file : ./models/v5Lite-g-sim-640.joint
image file : cengiz-sari-X4spr8Kuwxc-unsplash.jpg
img_h, img_w : 640 640
[AX_SYS_LOG] AX_SYS_Log2ConsoleThread_Start
Run-Joint Runtime version: 0.5.10
--------------------------------------
[INFO]: Virtual npu mode is 1_1

Tools version: 0.6.1.20
07305a6
run over: output len 3
--------------------------------------
Create handle took 492.77 ms (neu 34.07 ms, axe 0.00 ms, overhead 458.70 ms)
--------------------------------------
Repeat 10 times, avg time 22.56 ms, max_time 22.97 ms, min_time 22.48 ms
--------------------------------------
detection num: 18
 0:  94%, [1866, 1142, 2485, 2806], person
 0:  92%, [2417, 1240, 2971, 2807], person
 0:  89%, [1356, 1234, 1762, 2432], person
 2:  88%, [2827, 1334, 3797, 2230], car
 2:  85%, [3385, 1416, 4031, 2852], car
 0:  84%, [ 895, 1276, 1281, 2424], person
 0:  78%, [ 747, 1278,  926, 1729], person
 0:  77%, [  25, 1254,  213, 1809], person
 0:  73%, [ 419, 1325,  585, 1780], person
 0:  71%, [ 247, 1316,  423, 1801], person
28:  64%, [ 729, 1812,  998, 2319], suitcase
 0:  61%, [ 610, 1421,  744, 1729], person
 2:  53%, [3808, 1353, 4031, 1502], car
 2:  50%, [2782, 1353, 2954, 1519], car
 0:  42%, [1167, 1204, 1325, 1572], person
 0:  39%, [1318, 1261, 1459, 1632], person
12:  38%, [1861, 1370, 1949, 1530], parking meter
 0:  35%, [ 171, 1305,  284, 1788], person

opened by BUG1989 1

文档中我看4b可以运行10帧多？说是80ms左右，实际用cm4跑只有2帧，问题在哪呀

这是我的打印信息：

pi@dog:~/YOLOv5-Lite $ python detect.py --source 0 --img-size 320 /usr/local/lib/python3.9/dist-packages/torchvision/io/image.py:13: UserWarning: Failed to load image Python extension: warn(f"Failed to load image Python extension: {e}") Namespace(weights='weights/v5lite-s.pt', source='0', img_size=320, conf_thres=0.45, iou_thres=0.5, device='', view_img=False, save_txt=False, save_conf=False, nosave=False, classes=None, agnostic_nms=False, augment=False, update=False, project='runs/detect', name='exp', exist_ok=False) YOLOv5 🚀 v1.4-46-gcdf42dd torch 1.11.0 CPU

Fusing layers... /usr/local/lib/python3.9/dist-packages/torch/functional.py:568: UserWarning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument. (Triggered internally at /root/pytorch/aten/src/ATen/native/TensorShape.cpp:2227.) return _VF.meshgrid(tensors, **kwargs) # type: ignore[attr-defined] Model Summary: 230 layers, 1640661 parameters, 0 gradients, 3.9 GFLOPS 1/1: 0... success (320x240 at 30.00 FPS).

0: 256x320 Done. (0.336s) 0: 256x320 Done. (0.542s) 0: 256x320 Done. (0.508s) 0: 256x320 Done. (0.561s) 0: 256x320 Done. (0.494s) 0: 256x320 Done. (0.601s)

是用的v5lite-s模型 img-size也改成了320 请问还有哪里没设置对吗，为何帧数相差如此之大

opened by jd3096-mpy 2

Releases(v1.4)

v1.4(Mar 5, 2022)
update export.py to extract v5lite onnx model with concat head. @ppogg

add tensorrt inference sdk thanks for @ChaucerG

add onnxruntime inference sdk,thanks for @hpc203

add gcnet model , thanks for @315386775

undate yolo.py @ChaucerG @Alexsdfdfs @315386775

undate model.py @ppogg @Alexsdfdfs @315386775 Now YOLOv5-Lite support android, ncnn, mnn, tnn, onnxruntime, tensorrt, openvino, tflite. May be the repo will support more in the future~ Thanks for all the contributors of YOLOv5-Lite!

Source code(tar.gz)
Source code(zip)
YOLOv5-Lite-1.4.zip(3.49 MB)
v1.3(Oct 14, 2021)

add openvino demo add v5lite-c.pt add v5lite-c IR model link
Source code(tar.gz)
Source code(zip)
YOLOv5-Lite.zip(1.63 MB)
v1.2(Oct 14, 2021)

add v5lite-g.pt add mnn demo add model zoo link
Source code(tar.gz)
Source code(zip)
YOLOv5-Lite.zip(1.87 MB)

v1.1(Aug 25, 2021)

Remove some redundant code
Add the example of Android development
Release the first version of Android apk
Add lighter baseline
Add eval.py
Update baseline

# evaluate in 320×320：
Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.208
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.362
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.206
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.049
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.197
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.373
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.216
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.339
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.368
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.122
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.403
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.597

# evaluate in 416×416：
Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.244
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.413
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.246
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.076
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.244
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.401
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.238
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.380
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.412
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.181
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.448
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.626

# evaluate in 640×640：
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.271
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.457
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.274
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.125
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.297
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.364
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.254
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.422
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.460
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.272
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.497
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.616

Source code(tar.gz)
Source code(zip)
YOLOv5-Lite.zip(79.21 MB)

v1.0(Aug 24, 2021)

About shufflev2-yolov5: lighter, faster and easier to deploy. Evolved from yolov5 and the size of model is only 1.7M (int8) and 3.3M (fp16). It can reach 10+ FPS on the Raspberry Pi 4B when the input size is 320×320~
Source code(tar.gz)
Source code(zip)

🍅🍅🍅YOLOv5-Lite: lighter, faster and easier to deploy. Evolved from yolov5 and the size of model is only 1.7M (int8) and 3.3M (fp16). It can reach 10+ FPS on the Raspberry Pi 4B when the input size is 320×320~

Related tags

Overview

YOLOv5-Lite：lighter, faster and easier to deploy

Comparison of ablation experiment results

Comparison on different platforms

Detection effect

Base on YOLOv5

10FPS can be used with yolov5 on the Raspberry Pi with only 0.1T computing power

NCNN_Android_demo

More detailed explanation

Reference

Comments

YOLOv5 🚀 by Ultralytics, GPL-3.0 license

Parameters

YOLOv5 v6.0 backbone

[from, number, module, args]

YOLOv5 v6.0 head

repo

source code

model

result on AXera-Pi

Releases(v1.4)

v1.4(Mar 5, 2022)

v1.3(Oct 14, 2021)

v1.2(Oct 14, 2021)

v1.1(Aug 25, 2021)

v1.0(Aug 24, 2021)

Owner

pogg

This project aims to explore the deployment of Swin-Transformer based on TensorRT, including the test results of FP16 and INT8.

Based on Yolo's low-power, ultra-lightweight universal target detection algorithm, the parameter is only 250k, and the speed of the smart phone mobile terminal can reach ~300fps+

Yolov5-lite - Minimal PyTorch implementation of YOLOv5

LF-YOLO (Lighter and Faster YOLO) is used to detect defect of X-ray weld image.

KSAI Lite is a deep learning inference framework of kingsoft, based on tensorflow lite

A tutorial showing how to train, convert, and run TensorFlow Lite object detection models on Android devices, the Raspberry Pi, and more!

tensorrt int8 量化yolov5 4.0 onnx模型

EfficientNetv2 TensorRT int8

Code for "FPS-Net: A convolutional fusion network for large-scale LiDAR point cloud segmentation".

A platform for intelligent agent learning based on a 3D open-world FPS game developed by Inspir.AI.

A faster pytorch implementation of faster r-cnn

Learning to Reach Goals via Iterated Supervised Learning

A very simple tool for situations where optimization with onnx-simplifier would exceed the Protocol Buffers upper file size limit of 2GB, or simply to separate onnx files to any size you want.

Very simple NCHW and NHWC conversion tool for ONNX. Change to the specified input order for each and every input OP. Also, change the channel order of RGB and BGR. Simple Channel Converter for ONNX.

Multi-task yolov5 with detection and segmentation based on yolov5

TPH-YOLOv5: Improved YOLOv5 Based on Transformer Prediction Head for Object Detection on Drone-Captured Scenarios

Control-Raspberry-Pi-Robot-using-Hand-Gestures - A 4WD Robot car based on Raspberry Pi that controlled by hand gestures(using openCV and mediapipe)

DiffQ performs differentiable quantization using pseudo quantization noise. It can automatically tune the number of bits used per weight or group of weights, in order to achieve a given trade-off between model size and accuracy.

Python script for performing depth completion from sparse depth and rgb images using the msg_chn_wacv20. model in Tensorflow Lite.