zengyao1999

【队名】：真的很搞笑 【序号】：6 【论文】：Anchor-DETR-Query-Design-for-Transformer-Based-Object-Detection 【状态】：报名 【repo链接】：https://github.com/zengyao1999/Anchor-DETR-Query-Design-for-Transformer-Based-Object-Detection

我认为wrapbuffer_handle中并没有对vir_addr赋值，所以vir_addr为空指针，可能需要对dst_img.handle进行操作。不过我直接使用wrapbuffer_virtualaddr方法是能够正常转换的，只不过按照文档的说法，这样效率比较低

I have also trained my custom dataset with yolo-pose, maybe we can share the training process. If you change the number of key points, there are several changes to the...

@nicewiz First make sure your input is OK, including boxes and keypoints, and if you are not a single-class, modify nc and any other parameters involved. I modified the same...

@nicewiz How do you set the sigmas parameter for object keypoints similarity, is it custom? Or try to modify the formula of oks,like [this](https://github.com/TexasInstruments/edgeai-yolov5/pull/48/commits/d1f8421dbf6474a5b1dac500e0a6ef587ad3f557) ，in addition you can also see...

> > @nicewiz First make sure your input is OK, including boxes and keypoints, and if you are not a single-class, modify nc and any other parameters involved. I modified...

@SashaAlderson I used your method, the key point offset phenomenon disappeared, thank you very much!

> lkpt += kpt_loss_factor*（（1 - torch.exp（-d/（s*（4*sigmas**2）+1e-9）））*kpt_mask）.mean（） 它实际上会影响关键点的偏移量 > > > @SashaAlderson损失函数没有问题。我已经重新训练了模型，它们工作正常。我将在几天内更新模型。 工大大佬好！