ming71 / rotate-yolov3 Goto Github PK

RT

大佬你好，我在运行过程中遇到报错：
from utils.nms.r_nms import r_nms
ModuleNotFoundError: No module named 'utils.nms.r_nms
没有这个r_nms，请问是没上传这个文件吗？

请问各位大佬，进行训练时的hrsc.data的train和val路径是什么格式的内容？谢谢！

Hi Ming

We are interested in your good work based on the yolov3. We have tested this repo on the dataset HRSC2016 (all the images :1680) already. But, unfortunately, we cannot achieve that expected result-mAP around 80%. The tested results are as follows: P: 0.158, R: 0.413, [email protected]: 0.189, F1-score: 0.229.

The configurations are the same as the file provided in the package, hyp.py with 1800 epoches and yolov3-416 with such anchors 792, 2061, 3870, 6353, 9623, 15803 / 4.18, 6.48, 8.71 / -75, -60, -45, -30, -15 ,0,15, 30,45, 60,75, 90 (the same as the values in this file). Although these anchors are different for different datasets, they should not generate such effect on the experimental results obviously.

The model cannot detect the target in terms of most of detected images. We just show two cases with detected targets.
Hopefully we can obtain an expected performance after consulting.

Thanks in advance.

如题，我在进行其他yolo框架旋转框的尝试，现在进行了12天，感觉没什么进展，请问大概多久能完全改出来？

你好，我想请假一下你是怎么实现ORN的，ORN貌似不匹配pytorch1.1以上版本

您在repo中展现的舰船数据集，请问能提供下下载链接么？

Hello ming71, Iam farly new to the whole topik.
I trainet my own Yolo dataset, and now i want to go further. I dont understand axacly how i can customize yolo do i have to do it only in the config file or do I have to write new code, is there some nice explanation out there ? you can recoment ? Sory for opening an issue for this I did not knew how to contact you elswehre. Thank You

Hi, does backprop works on you skew iou calculation? This is because from i saw that you were using shapely in your calculation of skewiou which is not tensors

比如：mask = 144-215 anchors = 792, 2061, 3870, 6353, 9623, 15803 / 4.18, 6.48, 8.71 / -75, -60, -45, -30, -15 ,0,15, 30,45, 60,75, 90
我自己写了个倾斜版的anchor聚类，结果输出 9个 wha，比如对HRSC2016训练集进行聚类，得到
anchors: 153.60285000000002,29.688135000000017,-1.013242, 155.19675,29.394334999999984,0.6998453, 303.57180000000005,41.02582000000001,1.418342, 319.33545000000004,44.92217000000002,0.13387854999999999, 349.01929999999993,52.797129999999925,-0.7143881, 364.179,58.62256000000002,0.5916939, 397.41065000000003,66.58615,1.084582, 398.8361,69.37175500000006,-1.2166975, 533.6992,87.04564000000005,-0.2296711 Accuracy: 51.07%
请问如何转化为你的那种形式？

Hello，what is the format of the dataset？

Hi, ming71, Thanks for your repo, it is very wonderful.
when I training on custom dataset there is a problem, file train.py import test.py. and it import r_nms from utils.nms.r_nms. But No module named 'utils.nms.r_nms.
How can I get this file ? Thanks.

请问大佬，是直接回归xywh还有角度的吗，这个原理是啥子哦

Hi ming71, your work is very wonderful, when I training on ICDAR15 dataset there is a problem , which is "icdar_608_care.txt file is not exist in ${respo_root_dir}/utils/keans file dirctory " . How can I get this file ? Thanks.

大佬，UCAS_AOD数据集可以跑吗，map效果是多少，可以简单说一下吗，目前来说哪个效果最好可以说一下不

你好，我想问一下，你的旋转框的生成部分 具体是在代码的哪一个部分实现的？具体原理是？源代码看不太懂 ？

Hi @ming71
I'm also working with rotate object detection, I'd like to introduce you to an efficient way to get coordinates of a rotated box, please refer to get_corners.

Related to GIoU of 2 rotate boxes, did you find an efficient solution to calculate it? Can you please introduce it to me?

旋转检测模型打算支持Yolo v5 这个库吗？https://github.com/ultralytics/yolov5

ModuleNotFoundError:No module named 'utils.nms.r_nms', how to solve it? i can't find this moudle

在HRSC2016数据集上 的预训练文件可以分享一下吗？邮箱[email protected]感谢

RT

Thanks for your great work!

I was trying to train with my own dataset and I got below error.

File "A:\Object detection\rotate-yolov3-ming\rotate-yolov3-master\model\loss.py", line 245, in build_targets j[layer_id][(best_iou_indexes % na) * nt + gt_id] = True
TypeError: only integer tensors of a single element can be converted to an index

occurs from here:

I think below line should be changed to

layer_id = (best_iou_indexes//na)[0]

Is it right?

/content/drive/My Drive/rotate-yolov3/utils/nms
running install
running bdist_egg
running egg_info
writing r_nms.egg-info/PKG-INFO
writing dependency_links to r_nms.egg-info/dependency_links.txt
writing top-level names to r_nms.egg-info/top_level.txt
/usr/local/lib/python3.6/dist-packages/torch/utils/cpp_extension.py:335: UserWarning: Attempted to use ninja as the BuildExtension backend but we could not find ninja.. Falling back to using the slow distutils backend.
warnings.warn(msg.format('we could not find ninja.'))
writing manifest file 'r_nms.egg-info/SOURCES.txt'
installing library code to build/bdist.linux-x86_64/egg
running install_lib
running build_ext
building 'r_nms' extension
x86_64-linux-gnu-gcc -pthread -DNDEBUG -g -fwrapv -O2 -Wall -g -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2 -fPIC -I/usr/local/lib/python3.6/dist-packages/torch/include -I/usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include -I/usr/local/lib/python3.6/dist-packages/torch/include/TH -I/usr/local/lib/python3.6/dist-packages/torch/include/THC -I/usr/local/cuda/include -I/usr/include/python3.6m -c src/rotate_polygon_nms.cpp -o build/temp.linux-x86_64-3.6/src/rotate_polygon_nms.o -DTORCH_API_INCLUDE_EXTENSION_H -DTORCH_EXTENSION_NAME=r_nms -D_GLIBCXX_USE_CXX11_ABI=0 -std=c++14
In file included from /usr/local/lib/python3.6/dist-packages/torch/include/ATen/Parallel.h:149:0,
from /usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include/torch/utils.h:3,
from /usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include/torch/nn/cloneable.h:5,
from /usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include/torch/nn.h:3,
from /usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include/torch/all.h:7,
from /usr/local/lib/python3.6/dist-packages/torch/include/torch/extension.h:4,
from src/rotate_polygon_nms.cpp:1:
/usr/local/lib/python3.6/dist-packages/torch/include/ATen/ParallelOpenMP.h:84:0: warning: ignoring #pragma omp parallel [-Wunknown-pragmas]
#pragma omp parallel for if ((end - begin) >= grain_size)

src/rotate_polygon_nms.cpp: In function ‘at::Tensor r_nms(const at::Tensor&, float)’:
src/rotate_polygon_nms.cpp:3:39: warning: ‘at::DeprecatedTypeProperties& at::Tensor::type() const’ is deprecated: Tensor.type() is deprecated. Instead use Tensor.options(), which in many cases (e.g. in a constructor) is a drop-in replacement. If you were using data from type(), that is now available from Tensor itself, so instead of tensor.type().scalar_type(), use tensor.scalar_type() instead and instead of tensor.type().backend() use tensor.device(). [-Wdeprecated-declarations]
#define CHECK_CUDA(x) AT_CHECK(x.type().is_cuda(), #x, " must be a CUDAtensor ")
^
src/rotate_polygon_nms.cpp:8:3: note: in expansion of macro ‘CHECK_CUDA’
CHECK_CUDA(dets);
^~~~~~~~~~
In file included from /usr/local/lib/python3.6/dist-packages/torch/include/ATen/Tensor.h:3:0,
from /usr/local/lib/python3.6/dist-packages/torch/include/ATen/Context.h:4,
from /usr/local/lib/python3.6/dist-packages/torch/include/ATen/ATen.h:5,
from /usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include/torch/types.h:3,
from /usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include/torch/data/dataloader_options.h:4,
from /usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include/torch/data/dataloader/base.h:3,
from /usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include/torch/data/dataloader/stateful.h:3,
from /usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include/torch/data/dataloader.h:3,
from /usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include/torch/data.h:3,
from /usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include/torch/all.h:4,
from /usr/local/lib/python3.6/dist-packages/torch/include/torch/extension.h:4,
from src/rotate_polygon_nms.cpp:1:
/usr/local/lib/python3.6/dist-packages/torch/include/ATen/core/TensorBody.h:268:30: note: declared here
DeprecatedTypeProperties & type() const {
^~~~
src/rotate_polygon_nms.cpp:3:23: error: ‘AT_CHECK’ was not declared in this scope
#define CHECK_CUDA(x) AT_CHECK(x.type().is_cuda(), #x, " must be a CUDAtensor ")
^
src/rotate_polygon_nms.cpp:8:3: note: in expansion of macro ‘CHECK_CUDA’
CHECK_CUDA(dets);
^~~~~~~~~~
src/rotate_polygon_nms.cpp:3:23: note: suggested alternative: ‘DCHECK’
#define CHECK_CUDA(x) AT_CHECK(x.type().is_cuda(), #x, " must be a CUDAtensor ")
^
src/rotate_polygon_nms.cpp:8:3: note: in expansion of macro ‘CHECK_CUDA’
CHECK_CUDA(dets);
^~~~~~~~~~
error: command 'x86_64-linux-gnu-gcc' failed with exit status 1

I met this when i compile this program

i download the .pt model file and run with command "python detect.py --wights weights/without-matrix.pt --cfg cfg/***.cfg --source 2".. it doesnt work. i dont know which cfg file is correct. Can you show me a correct command with weights file and cfg file?

请问一下LZ，旋转框的数据标注格式是怎么样的，如果可以，对于完整的pipline，可否写一个类似于https://github.com/ultralytics/yolov3的pipline readme

Hi

I have two questions regarding the roated-yolov3

1. I saw the results achieved from kmeans.py,
   '''
   K anchors:
   [[153.87419355 31.30967742]
   [343.26412214 52.12366412]
   [574.57024793 109.7231405 ]]
   Accuracy: 69.19%
   -----area_cluster-----
   [[ 12130]
   [ 42951]
   [113378]]
   -----ratio_cluster-----
   4.18/1
   6.50/1
   8.75/1
   '''

but in yolov3-416.cfg, the anchor values: 792, 2061, 3870, 6353, 9623, 15803 / 4.18, 6.48, 8.71 / -75, -60, -45, -30, -15 ,0,15, 30,45, 60,75, 90. How did you obtain the first six value? area number : 6 but the ratio number: 3.

2. In terms of HRSC dataset, when using this repo, what  results did you get? such as 'P', 'R',  'mAP',  'F1' .  around 80%? 

thanks.

在编译make.sh时，出现了下面的错误，请您帮我看看，谢谢。

Excellent job!
I was trying to train the rotate-yolov3 model with HRSC2016 but the model seems to require a lower version of cuda. However I only have 3090 GPU which only support cuda 10+. I think you have trained it because of what you show in the Detection Results. So could you please provide the trained weight model on HRSC2016 dataset?

大佬 我按照你的思路改了个yolov4 但一直要过拟合 请问是什么原因

where i can get the weights file for testing?

感谢作者用yolo基于pytorch实现倾斜检测，其中丰富的注释对我有很大帮助。

本人目前运行环境已经搭建完毕，本来想跑通测试代码，但是由于作者提供的模型与配置文件不匹配暂时还未跑通。于是打算优先把训练代码跑通。我考虑是在作者提供的1张HRSC训练集上进行训练，然后再在该图片上测试结果。但是loss始终无法收敛，在1.5左右震荡。我目前只修改了配置文件中的学习率，初始为0.001，训练策略如下：
scheduler= lr_scheduler.MultiStepLR(optimizer, milestones=[100,200,300,400, 500, 600,700,800, 900], gamma=0.1)

目前有几个问题想向作者请教

• 在HRSC训练过程中还有哪些比较重要的参数需要调试

• README.md中展示的军舰检测效果图是旋转Yolov3训练后的测试结果展示图吗

• 是否可以提供一份预训练模型 HRSC+_raw_best.pt 的配置文件

期待可以在百忙中回复一下，祝顺意。

There is no a r_nms script. So cannot go through this package. Where is the r_nms script.

parser.add_argument('--cfg', type=str, default='cfg/ICDAR/yolov3_608_dh_o8_ga.cfg', help='cfg file path')
parser.add_argument('--data', type=str, default='data/tiny.data', help='*.data file path')

But if I want to test HRSC dataset, which one can I choose, yolov3-416.cfg or yolov3-m.cfg? thanks in advance.

大佬请问你是怎么样对先验框进行旋转处理的,在原始的yolov3中,有三种尺度的特征层进行预测,每一种尺度又对应3种不同大小的先验框,他们的大小比例是1:1,1:2,2:1左右,那么我觉得1:2和2:1应该属于同一种类别的先验框,只是他们的旋转角度不一样

你好！可以加你的微信进行沟通吗😭
最近在研究旋转框的目标检测，看到你的工作非常的优秀。

Hi, thanks for the effort.

Could you update readme with some basic steps how to use the scripts?
I'm getting "OSError: utils/kmeans/icdar_608_care.txt not found." error when running test.py, train.py and detect.py.

I am going through code, but some basic info would be very helpful.

Thanks

Hi Ming71,
This projects is a Creative idea, can you put this project in Ubuntu and Win running pipline, thank you very much. Looking forward to your reply.