- 博客(6)
- 问答 (1)
- 收藏
- 关注
原创 【代码bug】mmaction开源库之环境配置bug
【bug1】UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector. warnings.warn('Was asked to gather along dimension 0, but all 'fix: 该错误在有的服务器上使用pytorch 1.3版本时出现(不是所有机器都会有这个问题), 将版
2020-09-04 14:03:44 1410
原创 【自监督学习】supervised contrastive learning
摘要:Cross entropy is the most widely used loss function for supervised training of image classification models. In this paper, we propose a novel training methodology that consistently outperforms cr...
2020-08-28 10:12:01 3694
原创 【mmcv、mmdetection、mmaction】训练深度学习模型时如何自定义优化器“optimizer”,为模型不同layer设置不同的学习率
mmaction, mmdetection的代码库使用mmcv代码构建了深度学习训练框架,其中主要的类是’runner’ 。model和optimizer都是runner的成员。默认configs下,optimizer通过传入字典参数optimizer = dict(type=‘SGD’, lr=0.1, momentum=0.9, weight_decay=0.0005)进行初始化。此时,整个模型的学习率都是0.1,不可对复杂模型的特定层或者参数设定不同的学习率lr。本文展示了如何解决这一问题:
2020-08-26 19:41:17 2948
原创 【视频动作识别】I3D网络的代码细节
视频分析中常用的I3D代码细节目前写的比较好的两个video understanding base codemmactionslowfast这两个代码中对I3DI3DI3D 的实现方式分别如下#mmactoin# configs backbone=dict( type='ResNet_I3D', pretrained='modelzoo://re...
2019-12-23 23:50:36 4452
转载 【Personal Statement】5 Tips to Write a Brilliant Supporting Statement
A cover letter is something that’s become requested less frequently by recruiters in recent years. With many CVs now being submitted by email or via an online system, it’s rare you’ll be asked to subm...
2019-10-02 16:10:49 324
原创 【目标检测】refineDet算法的实现代码
refineDetmotivationone stage目标检测准确率低的原因是因为 class imbalance problem。已有的一些解决方案{object prior constraint[24]reshaping the standard cross entropy loss to focus on hard examples[28]max-out labeling m...
2019-05-13 18:15:43 884 1
空空如也
递归函数里面直接printf为什么结果就变了
2017-05-13
TA创建的收藏夹 TA关注的收藏夹
TA关注的人