ming71 / yolov3-polygon Goto Github PK

File "xx/RotationDetectors/othermethord/yolov3-polygon/models.py", line 254, in forward
metrics[0, i] = (self.losses['TP'][j] > 0).sum().float() # TP
IndexError: only integers, slices (:), ellipsis (...), None and long or byte Variables are valid indices (got numpy.float32)

What is the FPS you gain on gpu? thanks

自己创建了一个数据集，只有一个类别，按照教程改了相关参数设置训练时报错。但是把相关参数设置成两个类别的就没有错可以训练。报错如下：Traceback (most recent call last): File "/home/xuebq/Desktop/yolov3-polygon-master/train.py", line 182, in main(opt) File "/home/xuebq/Desktop/yolov3-polygon-master/train.py", line 117, in main loss = model(imgs.to(device), targets, requestPrecision=True) File "/home/xuebq/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/module.py", line 727, in _call_impl result = self.forward(*input, **kwargs) File "/home/xuebq/Desktop/yolov3-polygon-master/models.py", line 235, in forward x, *losses = module[0](x, targets, requestPrecision) File "/home/xuebq/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/module.py", line 727, in _call_impl result = self.forward(*input, **kwargs) File "/home/xuebq/Desktop/yolov3-polygon-master/models.py", line 148, in forward build_targets(pred_boxes, pred_conf, pred_cls, targets, self.scaled_anchors, self.nA, self.nC, nG, File "/home/xuebq/Desktop/yolov3-polygon-master/utils/utils.py", line 240, in build_targets tcls[b, a, gp_y_center, gp_x_center, tc] = 1 IndexError: index 1 is out of bounds for dimension 3 with size 1

您好，我看了代码。发现只预测四个顶点的坐标值，没有预测角度。所以这个代码只能检测水平框吗

how to convert coco-format instance segmentation annotations in Dota format?

I have tried the yolov3-tiny.cfg in ultralytics yolov3/cfg/, but it did have some mistakes.

作者您好，请问这个模型的算法不能保证输出框是矩形对吗？

您好，请问图片里这部分代码是在做什么工作呢？
xy = (xy @ M.T)[:, :2].reshape(n, 8)这行代码没有实现标签坐标的变化吗？

我看了下您的dataloader部分代码，是通过train.txt找到图片路径，然后通过对图片路径进行repalce得到标签的路径，我通过修改了一下dataloader直接读取ucas数据集，打开您代码里的plotFlag开关查看了label是否读取正确，也是正确读取了，但是训练的时候在第一个epoch中TP和FN一直为0，而FP一开始为十的四次方数量级，然后一直减少直到0，然后TP FN和FP就一直为0了。
所以我想请问一下您训练的标签的数据集格式是怎样的，根据你的代码，应该是class name x1 y1 x2 y2 x3 y3 x4 y4

作者：
你好，请问可以开源出1个权重么，我想看看具体啥效果。刚入坑旋转检测，还不太懂。

there is no trained models in your link(only detaset).Can you provide it?

there is some method convert lableme data to DOTA formate?

作者能否详细说下怎么训练类似ICDAR的自己的数据集

大佬你好，感谢你的无私开源，我想问下如何修改特征提取网络，我试过把yolov3.cfg改为v3_tiny.cfg,结果是报显存不足的错误

请问支持背景负样本训练吗？

Traceback (most recent call last):
File "E:/dl_pkg/yolov3-polygon-master/train.py", line 182, in
main(opt)
File "E:/dl_pkg/yolov3-polygon-master/train.py", line 117, in main
loss = model(imgs.to(device), targets, requestPrecision=True)
File "E:\Anaconda3\envs\torch1.3\lib\site-packages\torch\nn\modules\module.py", line 547, in call
result = self.forward(*input, **kwargs)
File "E:\dl_pkg\yolov3-polygon-master\models.py", line 237, in forward
x, *losses = module[0](x, targets, requestPrecision)
File "E:\Anaconda3\envs\torch1.3\lib\site-packages\torch\nn\modules\module.py", line 547, in call
result = self.forward(*input, **kwargs)
File "E:\dl_pkg\yolov3-polygon-master\models.py", line 151, in forward
requestPrecision)
File "E:\dl_pkg\yolov3-polygon-master\utils\utils.py", line 240, in build_targets
tcls[b, a, gp_y_center, gp_x_center, tc] = 1
IndexError: index 1 is out of bounds for dimension 3 with size 1

Process finished with exit code 1
请问这个是咋回事？，类别是1，数据集标注格式为cls,x1,y1,x2,y2,x3,y3,x4,y4,theta,x,y,w,h

楼主可以分享一下训练好的权重文件么？？

你好，请问一下用多卡训练出现这个问题可能是什么原因呢？

楼主您好，请问我增加batch_size报错是什么原因啊？ 谢谢
Traceback (most recent call last):
File "train.py", line 181, in
main(opt)
File "train.py", line 116, in main
loss = model(imgs.to(device), targets, requestPrecision=True)
File "/home/wen/anaconda3/envs/ryolo/lib/python3.7/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/wen/Documents/object_detection/yolov3-polygon/models.py", line 254, in forward
metrics[0, i] = (self.losses['TP'][j] > 0).sum().float() # TP
IndexError: only integers, slices (:), ellipsis (...), None and long or byte Variables are valid indices (got numpy.float32)

I've had great success implementing this on the UCAS-AOD dataset but am having issues trying other datasets.
I have my labels in the form class x1 y1 x2 y2 x3 y3 x4 y4 and have changed the .data, yolov3.cfg and .names files.
The most common error I encounter is when running train.py, it says "UnboundLocalError: local variable 's' referenced before assignment"
Is there anything else I should be changing?

Thank you!

Hello, thanks for your work.
I want to generate preset anchors for my own dataset , and my current label files are like 'cls x1 y1 x2 y2 x3 y3 x4 y4 ', how should I reorganize them?

你好，我想问你下，您这个旋转框实现的原理是啥？有啥论文依据吗？我看代码里面是最后回归了分别回归了四点坐标的loss，回归的坐标 有顺序吗？