spytensor / prepare_detection_dataset Goto Github PK

一幅图像中有多个目标，labels.csv 形式是/path/to/image,xmin,ymin,xmax,ymax,label, xmin2, ymin2, xmax2, ymax2, label2吗？还是需要分开两条记录，
/path/to/image,xmin,ymin,xmax,ymax,label
/path/to/image,xmin2,ymin2,xmax2,ymax2,label2

Thank you for your work. But your code is a little slow on the big data collection, and will you consider adding multithreading later?

Traceback (most recent call last):
File "labelme2coco.py", line 115, in
train_instance = l2c_train.to_coco(train_path)
File "labelme2coco.py", line 30, in to_coco
self.images.append(self._image(obj, json_path))
File "labelme2coco.py", line 58, in _image
img_x = utils.img_b64_to_arr(obj['imageData'])
File "/home/hustget/miniconda2/envs/open-mmlab/lib/python3.7/site-packages/labelme/utils/image.py", line 24, in img_b64_to_arr
img_data = base64.b64decode(img_b64)
File "/home/hustget/miniconda2/envs/open-mmlab/lib/python3.7/base64.py", line 87, in b64decode
return binascii.a2b_base64(s)
binascii.Error: Invalid base64-encoded string: number of data characters (73941) cannot be 1 more than a multiple of 4
这个ｂｕｇ需要怎么改

Hi, I found that the annotation id starts with 0 in the csv2coco.py file:

But for current official coco evaluation toolkit (https://github.com/cocodataset/cocoapi), it requires that the annotation id in ground truth file should start with 1, otherwise, it would cause wrong result. Check this: cocodataset/cocoapi#507 (comment)

Maybe we should change this to be compatible with coco evaluation toolkit.

{ "height": 2400, "width": 4000, "id": 2505, "file_name": "06266.jpg" }, { "height": 2400, "width": 4000, "id": 2506, "file_name": "06266.jpg" }, { "height": 2400, "width": 4000, "id": 2507, "file_name": "06266.jpg" }, { "height": 2400, "width": 4000, "id": 2508, "file_name": "06266.jpg" }, { "height": 2400, "width": 4000, "id": 2509, "file_name": "06266.jpg" }, { "height": 2400, "width": 4000, "id": 2510, "file_name": "06266.jpg" }, { "height": 2400, "width": 4000, "id": 2511, "file_name": "06266.jpg" }, { "height": 2400, "width": 4000, "id": 2512, "file_name": "06266.jpg" },

使用csv2coco转出来的json存在图片重复的情况，即一个box就生成了一张图，应该是多个box对应一张图才对

你好，请问下如果一张图片由多个目标，那就有多个bbox的坐标值，那csv的格式应该是下面(1)还是(2)呢：
(1) image_path1, x1min, y1min, x1max, y1max, cat1, x2min, y2min, x2max, y2max, cat2.....
(2) image path1, x1min, y1min, x1max, y1max, cat1
image_path1, x2min, y2min, x2max, y2max, cat2
.....

您好，我最近在做一个比赛，具体是mmdetection输出的bbox的json文件转成txt或xml格式，请问您写过这方面的程序吗，谢谢您

When csv2coco, only pictures with objections in csv file, how to convert pictures with no objection into coco datasets ?

我用您的代码转coco image字段的ID 不自增 可以交流以下吗 我VX wei4194162

１．使用labelme标注的时候要选择保存图像信息（Save with Image Data）;
２．labelme标注生成的.json需要与被标注的图片在同一目录。

如果一个图片中有多个框（这种情况很多见），那么csv格式是一行对应一个框还是一行对应一张图片呢？

可以做一个coco转voc数据集的tool吗？

你好，对于多个类别的情况，classname_to_id 中不止一个类别时，感觉是否存在一定问题？
因为这里，划分数据集是这样做的：
train_path, val_path = train_test_split(json_list_path, test_size=0.12)
期待大佬的回复，谢谢啦

csv转其他格式都有了，那么可以弄一个其他格式to csv的库吗？

annotation['area'] = 1.0

这里默认area=1.0生成的JSON 无法用COCO API统计APs APm APl

请问您的labelme2coco.py所对应识别的labelme版本号是多少？支持labelme4.5.6么？

how to convert coco format to voc

您好，我运行了您的代码，可是出现了

ValueError: With n_samples=0, test_size=0.15 and train_size=None, the resulting train set will be empty. Adjust any of the aforementioned parameters.

错误，请问train_target该设成什么呢

你好，请问下如果一张图片中有多个目标，那csv的一行对应的是一个目标框信息还是多个目标信息, 麻烦您有时间解答下，谢谢！