Open Hezhexi2002 opened 2 years ago
❔问题
在常规的yolo标签格式中,总共会有5列CLASS+xywh(归一化),我曾经在yolov5-face上训练,在面部关键点的末尾添加8个坐标(归一化)。所以我想弄清楚为什么转换后的标签中有 56 列?
附加上下文
我假设 56 列包括 34 个 17 kps 的坐标加上原点 CLASS+xywh(normalized) 但仍然有 56-(34+5)=17 列,所以我猜可能 17 kps 的坐标不仅包括 x, y 还有 z,在这种情况下,将有 17x3+5=56 列,与您在代码中提到的一样。但我不确定猜测是否正确,因为我之前没有了解人体姿势估计,我真的需要知道这一点,因为我想要我的自定义数据集,它只包含 4 kps,但有更多的类,而不仅仅是人。如果有人能给我一些建议,我将不胜感激。
每个关键点坐标后跟的有类别,0代表不存在,1代表遮挡,2代表存在
❔问题
在常规的yolo标签格式中,总共会有5列CLASS+xywh(归一化),我曾经在yolov5-face上训练,在面部关键点的末尾添加8个坐标(归一化)。所以我想弄清楚为什么转换后的标签中有 56 列?
附加上下文
我假设 56 列包括 34 个 17 kps 的坐标加上原点 CLASS+xywh(normalized) 但仍然有 56-(34+5)=17 列,所以我猜可能 17 kps 的坐标不仅包括 x, y 还有 z,在这种情况下,将有 17x3+5=56 列,与您在代码中提到的一样。但我不确定猜测是否正确,因为我之前没有了解人体姿势估计,我真的需要知道这一点,因为我想要我的自定义数据集,它只包含 4 kps,但有更多的类,而不仅仅是人。如果有人能给我一些建议,我将不胜感激。
每个关键点坐标后跟的有类别,0代表不存在,1代表遮挡,2代表存在
哦哦,原来如此,谢谢大佬,这样一来确实就是17*3+5=56列了,我再去看看代码:-)
❔问题
在常规的yolo标签格式中,总共会有5列CLASS+xywh(归一化),我曾经在yolov5-face上训练,在面部关键点的末尾添加8个坐标(归一化)。所以我想弄清楚为什么转换后的标签中有 56 列?
附加上下文
我假设 56 列包括 34 个 17 kps 的坐标加上原点 CLASS+xywh(normalized) 但仍然有 56-(34+5)=17 列,所以我猜可能 17 kps 的坐标不仅包括 x, y 还有 z,在这种情况下,将有 17x3+5=56 列,与您在代码中提到的一样。但我不确定猜测是否正确,因为我之前没有了解人体姿势估计,我真的需要知道这一点,因为我想要我的自定义数据集,它只包含 4 kps,但有更多的类,而不仅仅是人。如果有人能给我一些建议,我将不胜感激。
每个关键点坐标后跟的有类别,0代表不存在,1代表遮挡,2代表存在
您好,请问我在导出onnx模型的时候遇到了如下问题: 报错为: Warning: Constant folding - Only steps=1 can be constant folded for opset >= 10 onnx::Slice op. Constant folding not applied. Warning: Constant folding - Only steps=1 can be constant folded for opset >= 10 onnx::Slice op. Constant folding not applied. [W shape_type_inference.cpp:425] Warning: Constant folding in symbolic shape inference fails: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument index in method wrapper__index_select) (function ComputeConstantFolding) Warning: Constant folding - Only steps=1 can be constant folded for opset >= 10 onnx::Slice op. Constant folding not applied. Warning: Constant folding - Only steps=1 can be constant folded for opset >= 10 onnx::Slice op. Constant folding not applied. Warning: Constant folding - Only steps=1 can be constant folded for opset >= 10 onnx::Slice op. Constant folding not applied. ONNX: export failure: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument index in method wrapper__index_select)
请问您知道是什么原因导致的嘛?搜了很多教程,发现并不能解决我的问题,期待回复!
RectLabel is an offline image annotation tool for object detection and segmentation. With RectLabel, you can import the COCO keypoints format and export to the YOLO keypoints format.
class_index center_x center_y width height x1 y1 v1 x2 v2 y2 x3 y3 v3 ...
0 0.545230 0.616880 0.298794 0.766239 0.522073 0.309332 2 0.540170 0.293193 2 0.499589 0.296503 2 ...
A visibility flag v defined as v=0: not labeled (in which case x=y=0), v=1: labeled but not visible, and v=2: labeled and visible. https://cocodataset.org/#format-data
❔Question
In regular yolo label format there will be CLASS+xywh(normalized) 5 columns in total,and I used to train on yolov5-face which add 8 coordinates(normalized) in the end of the keypoints of face.So I want to figured out why there are 56 columns in your converted labels?
Additional context
I assume the 56 columns include 34 coordinates of 17 kps plus the origin CLASS+xywh(normalized) but there are still 56-(34+5)=17 columns,so I guess maybe the coordinates of 17 kps are not only include x,y but also the z,In that case,there will be 17x3+5=56 columns just same as you mentioned in your code.But I'm not sure the guess is right because I haven't learned about the human pose estimation before,I really need to know that because I want to my custom datasets which only contains 4 kps but with more classes instead of just person.I'd appreciate it if someone can give me some advice.