congjianting

train loss nan

7

@yangxue0827 你好, 我用自己的训练数据时会出现nan, 我的学习率是0.001, res50网络, 单卡, 这个现象是跟学习率有关吗? 你有什么好的建议解决下吗?

@JingChaoLiu

裁剪掉char特征效果能有多少?

1

@Real-YeJ 叶同学,你好, 实际工作中不太容易能够做到字符级别的标注或学习, 想问下, 如果不做字符级别的特征, 你们的模型效果大概在各项指标上是什么水平?

update readme.md about swin and convernext model?

1

@SuperHenry2333 是否可以更新下readme.md? modelzoo的模型和日志对应关系比较杂乱. 是否可以更新成类似res50的说明样式?

will DINO Code be implemented in mmdetection project?

4

@SlongLiu @FengLi-ust @IDEACVR thanks for your greate work.

onnx and trt infers different results in some shapes

2

## Description Hi, i found the results is different in some special shapes when using mmdet proj, the result is 10 pixels shifting in x direction. the solution is i...

hi, @pudae , what is your last loss value when the iou value = 0.346 in the val datasets?

@yutaocui 崔老师, 你好, 训练vit-base的模型报了如下一个warnnning, 麻烦您有空的时候看看是什么原因导致的? [W reducer.cpp:347] Warning: Grad strides do not match bucket view strides. This may indicate grad was not created according to the gradient layout contract, or...

跟踪目标的尺度变化剧烈如何调参?

6

各位老师好, 实际工作中的图片序列通常有尺度上的较大变化, 想问下这类情况除了search_scale_factor参数外 还有其它参数能做调整适应此类现象吗?

modelzoo not available!!!!!

1

hi, all models in modelzoo are not available.