BingfengYan

how to multi scale train using tf.data?

训练到中间会出现nan

15

你好，作者。非常感谢你的分享你的代码。我一直关注你的动态。从之前老版本的yolov3,到现在的版本。现在我用你的代码的时候，训练时，起初一开始就Nan,后来我把学习率放低，但是迭代到30几个epoch的时候就Nan了。我用你之前的版本并没有出现过这种情况。还有个奇怪的现象，我之前的学习率一般从0.001开始。现在0.0005，训到30还nan，这点有点奇怪

https://github.com/avBuffer/Yolov5_tf/blob/693a3af20043390350807b449b96d13212097d5f/core/yolov5.py#L498 ignore mask is iou (**all gt** and preds boxes) > thresh

https://github.com/YadiraF/face3d/blob/2fc26906d159a11398cd3e7a9b3f16b6f8937da3/face3d/mesh/cython/mesh_core.cpp#L41 这里应该判断在是否在3个顶点所在的线段上，不然会有个毛刺。 修改代码如下 if (dot00*dot11 - dot01 * dot01 == 0) { if (dot00 > dot11) { float u = v2.x / v0.x; float v = v2.y / v0.y; return...

how to train only one classes

4

hi, I want to train only one class, I edit num_classes=2, bug I get this error RuntimeError: Expected object of type torch.cuda.FloatTensor but found type torch.cuda.LongTensor for argument #3 'other'

see: https://github.com/BingfengYan/MOTSAM

You can find yolo with pose form this. https://github.com/BingfengYan/yolo_pose. train+test+to tensorRT

add CO_MOT

MSDeformAttn backward is a bit nondeterministic with floating point types. When trained multiple times, the final tracking performance varies greatly. Anyone Know Why, and How to improve

Update README.md

2

add VISAM